Intelligent Autonomous Systems 14

A computational theory model is constructed to generate a body map and a body schema. It is assumed that the fetus has a body map and a body schema to control and understand its body and generates them based on tactile information obtained by random motion in the womb called general movement. As a first step, in this research, a computational theory model is proposed to generate the body Map based on tactile information and to estimate nodes of the body schema by machine learning methods.

Recently, myoelectric interfaces have been intensively studied in various research fields. Because electromyography (EMG) is a bioelectrical signal, it can be influenced by many disturbing factors, e.g., electrode displacement, postural changes, and individual-dependent features like condition of muscles, subcutaneous fat, skin surface, etc., thus, it is difficult to realize high classification accuracy. To solve the problem, an EMG pattern classification method, which decomposes raw EMG signals into user/motion-dependent components by using a bilinear model, has been proposed. This enabled to reduce the time for classifier re-learning, however classification accuracy has not yet been sufficient. In the current study, we propose a signal decomposing method in consideration of the effect by forearm postures, in order to extract informative factors that correctly reflect hand gestures. We investigated the influences of postural changes exert on the classification accuracy of hand gestures, and tried to separate not only user dependent factor, but also posture-dependent component from EMG signals. As a result, we found that postural change decreases classification accuracy of approximately 20 % and we confirmed availability of our proposed method.

Bipedalism is one of the distinctive features of humans. However, humans in certain conditions use tripedalism for their locomotion. Patients who cannot bear their weight on their own legs or patients with balance disorders often use a cane. Temporal parameters have been defined for bipedalism, while they have not been defined for tripedalism. Therefore, in clinical rehabilitation, evaluation of patients’ gait using a cane is still very much a qualitative issue. In this study, we propose how we can define the temporal parameters for tripedalism. We calculated six quantitative numbers: (1) gait cycle of leg 1, (2) lag of foot strike between leg 1 and leg 2, (3) lag of foot strike between leg 1 and leg 3, (4) stance phase of leg 1, (5) stance phase of leg 2, and (6) stance phase of leg 3. With a set of these six elements, the foot strike and foot-off pattern of the three legs is uniquely determined. Because these elements are measurable quantitatively, we are able to express the tripedal gait in a quantitative manner with this number set. We call this number set “gait matrix”. The application of this gait matrix may be useful for evaluating patients’ gait using a cane in rehabilitation.

Maintenance of upright stance is one of the basic requirements in human daily life. Stance postural control is achieved based on multisensory inputs such as visual, vestibular and proprioceptive somatosensory inputs. In this paper, we proposed a stance postural control model including a neural controller with feed-forward inputs (muscle stiffness regulation) and sensory feedback of vestibular and proprioceptive somatosensory sensation. Through the optimization, variables of neural controller were designed to keep a musculoskeletal model standing during a 5 s forward dynamics simulation. From the results, we found that when both vestibular and proprioceptive somatosensory sensory input are available, low muscle stiffness is enough to maintain the balance of a musculoskeletal model in a stance posture. However, when vestibular sensory input get lost, higher muscle stiffness will be desired to keep the musculoskeletal model standing.

This paper describes a new method for estimating the body shape of a mobile robot by using sensory-motor information. In many biological systems, it is important to be able to estimate body shapes to allow it to appropriately behave in a complex environment. Humans and other animals can form their body image and determine actions based on their recognized body shape. However, conventional mobile robots have not had the ability to estimate body shape, and instead, developers have provided body shape information to the robots. In this paper, we describe a new method that enables a robot to obtain only subjective information, e.g., motor commands and distance sensor information, and automatically estimate its self-body shape. We call the method simultaneous localization, mapping, and self-body shape estimation (SLAM-SBE). The method is based on Bayesian statistics. In particular, the method is obtained by extending the simultaneous localization and mapping (SLAM) method. Experimental results show that a mobile robot can obtain a self-body shape image represented by an occupancy grid by using only its sensory-motor information (i.e., without any objective measurement of its body).

Although a posturography is commonly used for objective evaluation of static balance function, dynamic balance function is usually evaluated only with clinical scales. Simplified objective measurement systems for the evaluation of dynamic balance function need to be developed. In this preliminary study, we attempted to develop an index for the objective measurement of dynamic balance function from COP-COG data. The subjects comprised nine hemiparetic post-stroke patients and five healthy subjects. The simultaneous measurements of COG and COP were performed using a three-dimensional motion analysis system (Kinematracer, KisseiComtec, Japan) combined with force plate analysis. As indices for evaluating dynamic balance function, the latency of COP passing COG after heel contact (LCP) and the averaged |COP| − |COG| subtraction value during stance phase (ASV) were calculated. For evaluating validity of the measurement, the Berg Balance Scale, a frequently used clinical balance scale, was used. The results showed significant differences (0.13 ± 0.02 vs. 0.29 ± 0.23 s) between the healthy subjects and patients in LCP, and large, yet insignificant, differences (4.3 ± 0.5 vs. 2.7 ± 2.0 cm) in ASV. The ASV was strongly correlated with BBS. A strong correlation was observed between COG acceleration and ASV, except in one patient, who had a severe balance disorder. These results may encourage further investigation into the feasibility of COP-COG measurements for balance measurement.

This paper presents a master–slave finger exoskeleton developed to allow subjects whose brain activity is being measured by functional magnetic resonance imaging (fMRI) to remotely perform tasks. The MRI environment requires the device to be free from metal components and strongly immobilized, which can reduce the device’s versatility and ease of setup. To overcome these limitations, we designed a finger exoskeleton using pneumatic artificial muscles, which can be made metal–free and used for not only actuators but also sensors. We also proposed a symmetric, bilateral control method for the device, and experimentally validated device performance and its control method.

In daily lives, humans successfully transit their motions rather than performing separate movements. It has been widely acknowledged that there are four and five modules (called muscle synergy) in human sit-to-stand and walking motions, but it was still unclear how humans activate their redundant muscles to transit their movement from sitting to walking. Therefore this study hypothesize that human sit-to-stand can be explained from muscle synergies of sit-to-stand and walking motions, and we perform the experiment to verify it. Firstly, four and five muscle synergies were obtained from sit-to-stand and walking motion, and it has been tested whether these nine synergies are applicable to sit-to-walk motion. Results showed that sit-to-walk motion were successfully explained from nine synergies. Moreover, it was shown that humans adaptively changed the activation time of each synergies to delay body extension time and to generate necessary initial momentum for the walking motion.

In this paper we present a perception system for agriculture robotics that enables an unmanned ground vehicle (UGV) equipped with a multi spectral camera to automatically perform the crop/weed detection and classification tasks in real-time. Our approach exploits a pipeline that includes two different convolutional neural networks (CNNs) applied to the input RGB+near infra-red (NIR) images. A lightweight CNN is used to perform a fast and robust, pixel-wise, binary image segmentation, in order to extract the pixels that represent projections of 3D points that belong to green vegetation. A deeper CNN is then used to classify the extracted pixels between the crop and weed classes. A further important contribution of this work is a novel unsupervised dataset summarization algorithm that automatically selects from a large dataset the most informative subsets that better describe the original one. This enables to streamline and speed-up the manual dataset labeling process, otherwise extremely time consuming, while preserving good classification performance. Experiments performed on different datasets taken from a real farm robot confirm the effectiveness of our approach.

Planetary exploration scenarios illustrate the need for robots that are capable to operate in unknown environments without direct human interaction. Motivated by the DLR SpaceBot Cup 2015, where robots should explore a Mars-like environment, find and transport objects, take a soil sample, and perform assembly tasks, we developed autonomous capabilities for our mobile manipulation robot Momaro. The robot perceives and maps previously unknown, uneven terrain using a 3D laser scanner. We assess drivability and plan navigation for the omnidirectional drive. Using its four legs, Momaro adapts to the slope of the terrain. It perceives objects with cameras, estimates their pose, and manipulates them with its two arms autonomously. For specifying missions, monitoring mission progress, and on-the-fly reconfiguration, we developed suitable operator interfaces. With the developed system, our team NimbRo Explorer solved all tasks of the DLR SpaceBot Camp 2015.

The research at hand is part of the autonomous excavator project Thor. The long term-goal is the development of an excavator which is capable of performing landscaping tasks without human intervention. As this machine creates huge forces, safety plays an important role. This research extends the behavior-based trajectory generation for the excavation and truck loading operation, working for the undisturbed case only, with collision avoidance concerning obstacles and the machine itself. Based on a data efficient cylinder-coordinate octree for storing obstacles, joint movements are inhibited or actively influenced to prevent the machine from hitting objects and itself. Additionally, this research shows the suitability of the extensible reactive behavior-based control approach for the automation of construction machines.

Current operations in rail yards are dangerous and limited by the operational capabilities of humans being able to perform safely in harsh conditions while maintain high productivity. Such issues call out the need for robust and capable autonomous systems. In this paper, we outline one such autonomous solution for the railroad domain, capable of performing the brake bleeding inspection task in a hump yard. Towards that, we integrated a large form factor mobile robot (the Clearpath Grizzly) with an industrial manipulator arm (Yasakawa Motoman SIA20F) to effectively detect, identify and subsequently manipulate the brake lever under harsh outdoor environments. In this paper, we focus on the system design and the core algorithms necessary for reliable and repeatable system execution. To test our developed solution, we performed extensive field tests in a fully operational rail yard with randomly picked rail cars under day and night-time conditions. The results from the testing are promising and validate the feasibility of deploying an autonomous brake bleeding solution for railyards.

Conventionally, drones and robots are controlled using joysticks and remote controls. Here is a new idea: to approximate the feeling of “being the bird”, we introduce 3D fly-gestures and add VR-technology to stimulate our eyes. We chose a quadrotor and the combination of a Kinect, Virtual Reality Glasses and a Fuzzy-Controller as interface for processing instructions.

Localization and navigation inside GPS-denied buildings has been one of the main technological challenges of quadrocopter researches. Hereafter, this paper proposes and develops a supporting research platform integrated with 2D tag visual fiducials for quadrocopter indoor autonomous localization. Robot operating system (ROS) and Gazebo are simultaneously fused into the integrated platform. Under such circumstances, tag-involved images are sequentially captured via on-board cameras, while vehicle position/posture is achieved via off-board processing based on the open-source AprilTag algorithm. Simulation and experiments of the AR.Drone 2.0 are conducted to demonstrate the system architecture and workflow of the developed tag-in-loop indoor localization research platform. The results validate the effectiveness and application potentials of the ROS-Gazebo platform to support quadrocopters’ autonomous indoor localization, flight autopilot, and cooperative control, etc.

In the context of emergency scenarios, Unmanned Aerial Vehicles (UAVs) are extremely important instruments, in particular during monitoring tasks and in relation to the Post-Disaster assessment phase. The current paper describes a summary of the work performed during PRISMA [1], a project focused on the development and deployment of robots and autonomous systems able to operate in emergency scenarios, with a specific reference to monitoring and real-time intervention. Among other aspects, the investigation of strategies for mapping and for path following, for the implementation of Human-Swarm Interfaces and for the coverage of large areas have been performed, and they will be here summarized.

The use of quadrotors to civilian and military missions has been increased and the challenges involved on controlling it, mainly in indoors and restricted environments, has been attracting robotics researchers. Thus, an automatic collision avoidance approach is of utmost importance in this scenario, given the difficulty of control and the risk of accidents involved in the use of these vehicles. This paper presents an approach for obstacle avoidance of a manually controlled quadrotor automatically, allowing the operator to keep focus on the overall mission. The method is based on constantly estimating its future path considering its dynamics, current status, current control and distances measured by four on-board sonar sensors. Simultaneously, the pose is estimated based on the quadrotor odometry and an occupation grid representation of the nearby environment is constructed using the sonar sensors measurements. All that information is used to determine an imminent collision and overrides the user control, if necessary, keeping its last safe position. All the solution was evaluated in a simulator, the real quadrotor’s and sonars sensors were characterized to be embedded in the quadrotor through a computer-on-module and controlled over wireless network communication.

This paper presents a new control method that enable the robot pushes the obstacle while it moves to the target. In this paper, the shared control combines the user command—always directed at the target, and the machine command—to control the robot to avoid obstacles. And traditional obstacle avoidance requires to find a collision free path, which will find no solution in clutter environment. With pushing obstacles, the robot can solve a problem which was unsolvable, or find a more optimal solution (such as less motion for all joints). However, how to ensure the safety of pushing is a challenge. In this paper, an enhanced artificial potential field obstacle avoidance method is proposed, and a potential field for guaranteeing the safety of pushing is defined. The proposed method is test in simulation system, and the result shows that the proposed method is effective.

The ability to grasp is a fundamental requirement for service robots in order to perform meaningful tasks in ordinary environments. However, its robustness can be compromised by the inaccuracy (or lack) of tactile and proprioceptive sensing, especially in the presence of unforeseen slippage. As a solution, vision can be instrumental in detecting grasp errors. In this paper, we present an RGB-D visual application for discerning the success or failure in robot grasping of unknown objects, when a poor proprioceptive information and/or a deformable gripper without tactile information is used. The proposed application is divided into two stages: the visual gripper detection and recognition, and the grasping assessment (i.e. checking whether a grasping error has occurred). For that, three different visual cues are combined: colour, depth and edges. This development is supported by the experimental results on the Hobbit robot which is provided with an elastically deformable gripper.

In this paper, we study a method to manipulate objects in position-controlled robot hands: in-hand caging manipulation. In this method, an object is caged by a hand throughout manipulation and located around a goal as a result of the deformation of the cage without sensing the object configuration. A motion planning algorithm for planar in-hand caging manipulation of a circular object is proposed. Planned motions by the algorithm are successfully performed by a two-fingered robot hand. Additionally some examples of 3D in-hand caging manipulation by a four-fingered robot hand are presented.

This paper reports on design and development of new cosmetic gloves for Myoelectric Prosthetic Hand which provides a realistic appearance and flexible motion of robot hands. The main design issues are divided into five as followings; appearance, gripping performance, durability, texture, flexibility. The appearance includes the shape, wrinkles, finger mark, nail, and color of the hand; the aim is to make these properties of the prosthetic hand as similar to those of the human hand as possible. The durability is evaluated by adaptabilities for daily living, and flexible materials without prevention from finger motion. Furthermore, the gripping performance is improved by the thickness map of palm which is well fit to the gripping object. The experiment shows the results of the performance test applied to the pick-and-place task by using powered prosthetic hand in order to evaluate total properties of the developed cosmetic gloves.

This paper proposes an Open Semantic Framework for knowledge acquisition of cognitive robots performing manipulation tasks. It integrates a Cloud-based Engine, which extracts discriminative features from the objects and generates their manipulation actions, and an Ontology, where the Engine saves data for future accesses. The Engine offloads robots by transferring computation on the Cloud. The Ontology favors knowledge sharing among manipulator robots by defining a common manipulation vocabulary. It extends the work proposed by the IEEE RAS Ontology for Robotics and Automation Working Group by covering the manipulation task domain. During ontological data insertion, data duplication is avoided by providing a novel efficient interlinking algorithm. During their retrieval, visual data processing is optimized by using a cascade hashing algorithm that intelligently accesses data. No training is required for object recognition and manipulation because of the adoption of a human-robot cooperation. The framework is based on the open-source Robot Operating System.

Human-following has become one of the most import functions as to human-friendly robots that are able to coexist with humans and serve humans. In this paper, we propose a tracking control strategy for a human-following robot which applies a sonar ring as its rangefinder to detect human and obstacles. In order to detect human, we equip the human with a guiding device. The tracking control strategy consists of firing strategy and human-following and obstacle-avoiding strategy. Firing strategy can reduce crosstalk effectively. Human-following and obstacle-avoiding strategy ensure that the robot can follow human in a constant following distance with an appropriate orientation and avoiding obstacles under unknown environments.

It has been shown that mobile robots could be a potential solution to home bio-monitoring for the elderly. Through our previous studies, a mobile robot system that is able to recognize daily living activities of a target person has been developed. However, in a home environment, there are several factors of uncertainty, such as confusion with surrounding objects, occlusion by furniture, etc. Thus, the features extracted could not guarantee the correct recognition. To solve the problem, we applied active sensing strategy to the robot, especially to the body contour based behavior recognition part, by implementing 3 algorithms in a row, which enabled (1) judging irregularity of feature extraction; (2) adjusting robot viewpoints accordingly; (3) avoiding excessive viewpoint adjustment based on a short-term memory mechanism, respectively. As a result of experiment in a home living scenario, higher activity recognition accuracy was achieved by the proposed active sensing algorithms.

When studying humanoid robots, many of the processes involving the synthesis of robots motion have much in common with problems found in biomechanics and human motor-control research. Based on the use of simulation tools, additive manufacturing and system identification techniques, this work presents the design and implementation of an underactuated biped robot guided via elastic elements. In particular, the focus was on the analysis of stable posture maintenance during standing obtained on a completely passive humanoid robotic system. An estimation of the mechanical parameters of the elastic mechanical network was performed using an Extended Kalman Filter. The results provided here are part of a larger project that aims to implement a hybrid robotic system where the balance of bipeds are obtained using passive elastic elements, thus simplifying the control of gait. This work contributes to the research and development of increasingly efficient human-like robots.

The complex of human foot plays an important role in the locomotion. Properly replicating the human foot characteristics on the humanoid robot foot design is supposed to improve the robot locomotion performance. In this research, we proposed three kinds of foot design, stiff foot, the windlass mechanism foot (with stiff plantar fascia) and the windlass mechanism foot (with elastic plantar fascia). Using a musculoskeletal biped robot and via a large set of dropping jump experiments, we confirmed that (1) the robot could achieve toe-off motion in the lifting off phase of jumping and (2) the windlass mechanism could increase the jumping height. This investigation on robot foot is expected to both improve the humanoid robot jumping performance and help us understanding how human achieve high performance locomotion.

This paper describes a motion planning architecture for a quadruped robot used for conveyance tasks. The architecture is designed to guarantee robot’s locomotion stability using only proprioceptive information. The robot is equipped with linear displacement sensors and pressure sensors located on the hydraulic cylinders used as actuators, a gyroscope on the main body and contact sensors on the leg tips. The robot knowledge is limited to proprioceptive sensory data and no a priori information is given about the environment. The resulting walking behaviour is validated through simulations on both flat terrain (with unknown objects along the path) and slopes. The gait is performed using a generic leg sequence and a simplified foothold planner. Initial experiments on the real platform have been carried out as well.

A heuristic path following algorithm for wheeled mobile robots is presented. This approach is based on the orthogonal projection to the path and exponential functions for lateral and longitudinal control. It allows smooth and stable navigation in dynamic and cluttered environments, and does not depend on the robot’s kinematics. The results are experimentally demonstrated using three different kinematic configurations: omnidirectional, Ackermann- and two-steering.

The aim of this paper is to face with the problem of localizing a robot during the navigation in a partially unknown environment. This feature becomes particularly noteworthy especially in the case of a colony of robots, possibly working with humans, inside a scenario where motion issues are crucial. Within this context the focus on self-localization through GPS and INS/SINS integration overtakes merely questions about algorithm efficiency because self-localization is a relevant part of the task. Thus, unlike other approaches, we have focalised on this behavior as an attitude an autonomous system should enhance during the task execution. The tight coupling of GPS and INS sensors is understood as a mechanism which provides the autonomous robot with a refinement of INS use by comparing and/or adjusting the INS performance by exploiting the GPS-INS integration.

For navigation of mobile robots in real-world scenarios, accurate and robust localization is a fundamental requirement. In this work we present an efficient localization approach based on adaptive Monte Carlo Localization (AMCL) for large-scale indoor navigation, using vector-based CAD floor plans. The approach is able to use the line segment data of these plans directly. In order to minimize the computational effort, a visibility lookup table is generated, reducing the amount of line segments to process for pose estimation. In addition, we show that the proposed approach performs well in cluttered as well as uncluttered environments. It is compared with grid map-based AMCL and is able to improve its results in terms of memory usage and accuracy.

This paper presents a virtual force guidance law for trajectory tracking of autonomous vehicles. Normally, three virtual forces are designed to govern the vehicles. The virtual centripetal force counteracts the influence of the reference heading rate. The virtual spring force pulls the vehicle to the reference trajectory and the virtual drag force prevents oscillations. When local obstacles are detected, an extra virtual repulsive force is designed to push the vehicle away from its way to get around the obstacles. Using the guidance law, the reference trajectory can be straight line, circle and general curve with time-varying curvature. The guidance law is directly applicable to path-following problem by redefining the reference point. The use of artificial physics makes the guidance law be founded on solid physical theory and computationally simple. Besides, the physical meanings of the parameters are definite, which makes it easy to tune in application. Simulation results demonstrate the effectiveness of the proposed guidance law for problems of trajectory tracking, path following, and obstacle avoidance.

Path planning in complicated 3D environments with narrow passages and rooms is a challenging problem, which is usually time consuming for geometric searching methods or incomplete for sampling-based methods. Focusing on these issues, this paper presents a new algorithm named 2D Voronoi-based Random Tree which combines the completeness of the voronoi diagram based 2D path searching methods and the efficiency of sampling based path planning methods. In this method, 2D voronoi diagram is created and used to guide the growth of random trees in complicated 3D environments. In each iteration of random trees growth, a new node on voronoi edges is selected by moving forward to the target with a fix step length along edges. And then, the 3D particles are distributed locally around this node and be selected according to a cost function for the random trees growth. By doing so, this method can find a valid path in complicated 3D environments with narrow passages and rooms while improving its efficiency and completeness. To demonstrate its effectiveness, efficiency and robustness, the proposed method is examined and compared with RRT algorithm in various practical complex 3D environments.

The FieldLut algorithm is a widely used localization algorithm in the RoboCup MSL (Middle Size League). It is now used for indoor mobile robot localization, but it can only use 2D range data. This paper improves the FieldLut algorithm to allow the use of 3D range data for indoor localization and uses Kinect sensor as the input sensor. The core of our improvement is the creation of a 3D LUT (lookup table). The 3D LUT is created as a multi-layer 2D LUT. Additionally, a memory optimization method is proposed. Experimental result shows real-time performance at video rates and high accuracy; for example, using Kinect sensor, the localization error is below 15 cm in a 13 × 8 m room and the repeat localization is below 6 cm.

Accurate image fusion is an essential technique to obtain more information from remote sensing image in different sensors. This paper presents a method for fusion delineating objects from multiple sensors. The proposed algorithm partitions feature-based mutual information into the maximization as the requirement for fusion, which consists of entropy in the image. The wavelet transform decomposes the maximum value of the mutual information for image fusion. To evaluate the validity of the proposed method, experiments were conducted using two types of remote sensing images. The overlapping, correctness, and quality of the fusion object are over 98 %, 95.3 %, and 95.1 % respectively, which proves the proposed method is a promising solution for registration and fusion from two remote sensing images.

The information of the number of passengers getting in/off a vehicle is very important for public bus transport companies. In fact, the operators need to estimate the number of travellers using their vehicles for marketing purposes, for evaluating transit service capacities and allocating the proper number of buses for each connection-line. The goal of this work is to provide a system for counting and monitoring passengers, both adults and children, at the entrance of bus. This system is mainly based on an RGB-D sensor, located over each bus door, and image processing and understanding software. The RGB image could be affected by a high luminescence sensibility, whereas depth data allow a greater reliability and accuracy in people counting. The correctness and effectiveness of our method has been confirmed by experiments conducted in a real scenario. Furthermore, this approach has the advantage of being computationally inexpensive and flexible enough to obtain, in real time, statistical measures on the amount of people present in the bus, with the use of an Analytical Processing System (a separate process) that accesses the data stored in the database and extracts statistical data and knowledge about the bus passengers.

In this work, we present a software library which enables the efficient use of the Kinect One, a time-of-flight RGB-D sensor, with the nVidia Jetson TK1, an ARM-based embedded system, for the purpose of people detection. Our software exploits nVidia CUDA to process all data necessary for robust people detection algorithm and other perception algorithms by parallelizing the generation of the 3D point cloud and many pixel-wise operations on both the raw depth and the infrared images coming from the Kinect One sensor. The library developed has been released as open-source and the whole system has been tested as a people detection node in an open source multi-node RGB-D tracking framework (OpenPTrack). The results gathered show that the proposed system can be effectively used as a people detection node, outperforming the state-of-the-art in terms of people detection frame rate not only with the nVidia Jetson, but also with non-embedded computers.

Due to the low training time and high time resolution, steady state visual evoked potential (SSVEP)-based brain computer interfaces (BCIs) have been largely studied in recent years. The stimulus properties such as frequency, color and shape can greatly affect the performance, comfort and safety of the brain computer interfaces. Despite this fact, stimulation properties have received fairly little attention. This study aims to investigate the influence of stimulus color on SSVEPs. We tested 10 colors and did evaluation by using three different methods: the power amplitude, multivariate synchronization index, and canonical correlation analysis. The results showed that violet color had the least influence to SSVEPs while red color tended to have stronger impact.

A PSD (Position Sensitive Detector) -based microscopy was introduced previously by same authors to scan sophisticated objects using the PSD sensor and a laser beam connected to an X-Y table (Rahimi et al., 2014 IEEE International Conference on Robotics and Biomimetics (ROBIO), pp 1685–1690 (2014) [1]; Rahimi et al., 2015 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp 4955–4960 (2015) [2]). The system has numerous capabilities including but not limited to tracing an unknown object on the PSD, finding the dimensions of the object and adaptive local scanning for objects with intersections and bifurcations. Although promising results were obtained from that system, the speed of the scanning was still dependent on the speed of the movement of the micromanipulator. This work presents an extension to the tracing method by scanning tree-shaped objects. The simulations show a very successful scanning that can map any tree-shaped object.

The main function of a hand is to grasp and manipulate objects, and the thumb contributes most to this function. However, thumb rehabilitation devices, especially soft robotic gloves, have not been widely investigated. Soft pneumatic actuators are lighter, more flexible, and easier maintenance than other actuators. This makes them safer and more cost-effective. In this paper, we present a design for a soft robotic thumb rehabilitation system. We used a soft pneumatic actuator for bending (SPAB) for flexion-extension motion. For the carpometacarpal (CMC) joint, we used a parallel link mechanism by placing pneumatic artificial muscles (PAM) around it in three directions. We evaluated whether this setup of SPAB and PAM could enable the required three-dimensional thumb motions by using a prototype system on a dummy thumb. The results showed that the proposed mechanism enables opposition, abduction, and adduction motions for the thumb.

Hemiplegia is one of the major deficits caused by strokes. Mirror therapy (MT), which causes visual illusion to severely impaired side through presenting mirror images of the normal side in exercises, has been used in rehabilitation practice. However, due to individual differences, it is difficult for some patients to experience visual illusion. We had a hypothesis: vibration simulation for kinesthetic illusion, synchronized with visual stimulation with mirrored images could improve the elicitation of visual illusion. In this research, functional magnetic resonance imaging (fMRI) was employed to test the hypothesis. The brain activities of the subjects performing finger motor tasks with MT and vibration stimulation with the parameters identified in another experiment were compared. As a result, it was found that the parietal lobe of both sides of most of subjects showed higher activities. These two facts suggested MT synchronized tactile stimulation could achieve a higher possibility of visual illusion.

This research aims to develop a robot to help nursing students learn how to physically transfer paralyzed patients. Our prior robotic prototype had only been designed to imitate a patient with weak lower limbs; it was unable to imitate the imbalance and instability of the trunk. Therefore, we developed and waist joint on the prototype robot and also the control system in order to emulate quadriplegic and hemiplegic patients’ tendency to fall over. The waist was designed with 2 DOF compliant joints. Evaluation of the robot showed that the robot was able to properly imitate the unstable waist movements of paralyzed patients.

Modular robotic systems consists of more than one modules that are homogeneous and this paper proposes a novel structure design of modular robots called M-Lattice, which stands out for its load-capacity and high efficiency in self-reconfiguration. Compressibility of M-Lattice is guaranteed by the topological structure and the amount of compression has been calculated. The motion space indicates that different types of motion can be accomplished without interference. Two prototypes have been made to verify the motion ability of the modules and experiment shows the motion process of M-Lattice system.

The adaptive synchronized formation control problem of multiple mobile robots is studied in this paper. The communication constraints, including time-varying delays and data sampling, are considered in problem formulation and system design. Furthermore, the parameter uncertainties in system dynamics have also been taken into account and an adaptive formation controller is presented which enables the robot network to achieve the synchronized formation task adaptively. Convergence analyses of the proposed method are presented and several useful properties are provided. Simulation results validate the effectiveness of the proposed adaptive synchronized formation control approach.

When multiple robots cooperate to carry out some tasks, the movement synchronization of the formation is essential. However, due to the uncertainty of the complicated environment, it is inevitable that certain robot fails. Therefore, it’s worth to figure out some ways to minimize the damage caused by failed robot. In this paper, a distributed self-healing algorithm for multi-robot network global optimal movement synchronization is presented. The proposed algorithm can transform any robot’s failure into the failure of the robot which has the least degree so that can minimize the influence on movement synchronization. In addition, gradient is involved in the algorithm to make repairing path shortest under the condition of maintaining global optimal movement synchronization. In the whole process of the self-healing, all robots only communicate with their own neighbors, so the robots are distributed controlled. Finally, the effectiveness of the proposed algorithm is validated by simulation experiments.

In this paper, we investigate the consensus problem of multi agent systems based on the discrete-time general linear model with actuator saturation. A new family of distributed parametric low-and-high gain control protocols is provided, and such control protocols rely on the discrete-time Parametric Algebraic Riccati Equation. Furthermore, the convergence results are also given as consensus function sequences. Finally, an illustrative example shows that the low-and-high gain control protocol is effective for the consensus problem of the general linear multi agent systems with actuator saturation.

In this paper, we model topology-based partitioning strategies for multi-robot area coverage with no communication as a balanced graph partitioning problem, which is subject to connectivity and reliability requirements. We formalise it as a combinatorial optimisation problem, propose a generalised formulation that encompasses its most relevant features, and we introduce two variants of the basic problem. Furthermore, we discuss a metaheuristic solution method based on a hybrid Ant Colony Optimisation approach. Finally, the performance of the solution is validated using simulations.

We present a novel decentralized approach to allocate agents to tasks whose costs increase over time. Our model accounts for both the natural growth of the tasks and the effort of the agents at containing such growth. The objective is to minimize the increase in task costs. We show how a distributed coordination algorithm, which is based on max-sum, can be formulated to include costs of tasks that grow over time. Considering growing costs enables our approach to solve a wider range of problems than existing methods. We compare our approach against state-of-the-art methods in both a simple simulation and RoboCup Rescue simulation.

This paper presents a combined feedforward plus feedback control approach to compensate the hysteresis effect, which degrades the positioning accuracy of piezo-actuated mechanism. The LuGre friction model is extended to represent the nonlinear dynamics of the piezo-actuated positioning mechanism, and then the unknown model parameters are identified with the particle swarm optimization (PSO). Based on the developed mathematical model, the inverse LuGre model based feedforward plus feedback control is designed for the motion tracking control. Experimental results show that the LuGre model based hybrid control approach achieves a satisfactory position tracking performance. Owing to a simple structure, the proposed control approach can be implemented in other types of hysteretic systems.

For the Optimal tracking problem of a class affine non-linear system, based on discrete-time State Dependent Riccati Equation(DSDRE), an optimal tracking sample-data control method is proposed. Moreover, the optimization is discussed with relation between HJBE and DSDRE; stability of close-loop system is investigated with Lyapunov theorem. Finally, the method is the verified in numerical simulations.

In this paper an on-line track filtering and association algorithm for flight test was proposed. Firstly, a K-means clustering based scheme was used for track initialization and initial state and corresponding covariance matrix estimation for second-order extended Kalman filter. After that, track filtering and association and frequency estimation of Dutch roll were achieved through interactive use of the second-order extended Kalman-filter and the Kalman-predictor based QDA minimum error rate Bayesian classifier. Experimental results had shown that the algorithm can initialize the track reliably, filter and associate tracks in real time and estimate the frequency of flight testing precisely.

This paper uses a Smith predictor or time delay compensation in remote control system of mobile robot. Time delay causes a few errors, and the error in turns influence the control of the target. Smith predictor requires an accurate model, the type of a virtual model with a time delay in the feedback part, to compensate for the error value for the time delay. In the experiment of this paper, the error of Smith predictor is compensated by using a sliding mode controller. The mobile robot is implemented through the simulation that imaginary modeling is composed of two drive-wheels and one fixed-wheel. The mobile robot is controlled by sliding mode controller.

This paper aims to solve synthesis problems of two-wheel difference speed mobile robot in navigation and localization. A new method is proposed by combining the A* algorithm, motion control, and dead reckoning before the integrated treatment to address practical problems of short-term navigation and orientation during robot movement. First, the A* algorithm is used to plan out robot’s trajectory from a known point to the target, Floyd algorithm and circular arc algorithm are chosen for the optimization of the A* algorithm to make it in accordance with the trajectory of two-wheel differential motion robot. Then, the robot is controlled to move along the scheduled trajectory. Finally, based on dead-reckoning, robot’s position in the environment during locomotion is calculated to realize real-time positioning and tracking of its position and trajectory.

Aiming at space transposition using Space Robotic Arm (SRA), flexible docking between SRA’s end effecter (EE) and grapple fixture (GF) is the most important for space tasks. To avoid position errors leading to large contact force between EE and GF in the docking process, an adaptive impedance control method is proposed in this paper. PID feedforward with adaptive parameters is added into the impedance controller, and the force error function is used to deduce the adaptive parameters according to Lyapunov stability theory, which makes the force error decrease automatically during the connection process. Simulation proves that the adaptive impedance strategy gets better force control effect than the traditional impedance algorithm. Finally the SRA EE/GF connection experiments were conducted respectively based on traditional and adaptive impedance control strategy. The results showed that the adaptive impedance control strategy can achieve better control effect than the traditional strategy.

The hardware-in-the-loop (HIL) simulation (also called hybrid simulation) is a useful and flexible approach for the simulation of contact dynamics in space. In this study, a HIL contact simulation system including the motion simulator, control system and algorithm is introduced. The simulation divergence problem due to the time delay is studied. To compensate the time delay, the phase lead method is used compensate the force measurement delay, and the response error based force compensation is used to compensate the dynamic response delay of the motion simulator. The compensation requires the force measurement delay value, but does not require the dynamic response model. The effectiveness of the HIL simulation system and the delay compensation approach are verified by the simulations and experiments.

We propose a novel method, a particle filter on episode, for decision makings of agents in the real world. This method is used for simulating behavioral experiments of rodents as a workable model, and for decision making of actual robots. Recent studies on neuroscience suggest that hippocampus and its surroundings in brains of mammals are related to solve navigation problems, which are also essential in robotics. The hippocampus also handle memories and some parts of a brain utilize them for decision. The particle filter gives a calculation model of decision making based on memories. In this paper, we have verified that this method learns two kinds of tasks that have been frequently examined in behavioral experiments of rodents. Though the tasks have been different in character from each other, the algorithm has been able to make an actual robot take appropriate behavior in the both tasks with an identical parameter set.

The strength of carbon fibre parts depends on the fibre arrangements all over them, but only manual and sparse checks are usually executed to assess their quality. Here, we present an automatic method for computing the fibre orientations in each part point and mapping them onto the 3D model of the part. This process is automated by a robot that moves the measurement sensor above the object to be scanned. Since this sensor needs to acquire multiple images of the same point with different illuminations for correctly estimating the fibre orientation, we developed algorithms for online image registration in presence of translational sensor motion. Moreover, we propose real-time methods for projection of the estimated orientation vectors to a 3D model. Experiments show that this software allows the accurate and fast mapping of carbon fibre parts by means of an industrial robot. Accuracy assessments report a measurement accuracy below 5$$^{\circ }$$.

The special fiber bundle image has characteristics of anti-radiation and resistance to high temperature, thus can be used to observe nuclear environment. But there are honeycomb artifacts in fiber bundle image, and the resolution of image is low, which greatly affects the image quality. This paper propose a method to segment optical fiber cladding and fiber center. And an improved non local means (NLM) algorithm is used to denoise and repair the fiber bundle images. The algorithm can remove the honeycomb artifacts and greatly enhance the image resolution. The test results show that the algorithm is effective.

When water strider robots row on water, the periodically stroking water surface of the actuating legs will unavoidably bring vibrations and instabilities that might cause the robots to sink into water. In this work, a stability analysis model for water strider robots rowing on water was proposed and a mass-spring-damper-like model was defined to describe the robot-water interactions. We applied this model to evaluate the water-surface stability of a miniature surface tension-driven water strider robot by detaily discussing the effects of the actuating legs’ rowing with different rowing frequencies on the vibration, pitching and swinging motions. The theoretical results indicates the robot possesses a good water-surface stability. The stability analysis model presented in this study can help with the design of water strider robots in future.

In this study, we investigated a mechanism that allows a mobile robot to apply a large force to the environment. We first investigated the limits on the force that a mobile robot can apply to a target object by analyzing the forces between the robot, ground, and object and the limits on the frictional forces between them. To prevent the mobile robot from falling when applying a large force, we developed a prototype in which the manipulator was connected via a passive rotational joint. We investigated the pushing capacity of the prototype robot through an experiment in which it tilted a large object. The results confirmed that the mechanism allows a mobile robot to apply a large force to an object without falling by trial and error.

Robots on home environment have to deal with their environment that changes every time they perform tasks. In order to reduce improving descriptions for task and action planning manually, we propose a new framework to use probabilistic graphical model, which enables robot agent know which action in the task affects the result of the task the most, and by improving parameters of the action robot can maintain high success rate of tasks. Our framework let robot infer failure action of tasks using data from early task performance which are automatically recorded and retrieved with high-level data retrieval query interface. We evaluated our approach using mobile manipulation robot PR2 on daily assistive environment and task.

We study a novel robot programming method that uses the view-based approach: “view-based teaching/playback.” This method directly uses images for robot programming and can accommodate itself to changes of task conditions. However, our previous view-based teaching/playback cannot perform force-control tasks; for example, it cannot deal with pressing objects against walls, in which view of images does not change. In this paper, we extend the view-based teaching/playback so that it is applicable to force-control tasks using photoelasticity. In the experiment, the extended view-based teaching/playback succeeded in wall-pressing tasks.

Supervised machine learning techniques have proven very effective to solve the problems arising from model learning in robotics. A significant limitation of such approaches is that internal models learned for a specific robot are likely to fail when transferred to a robot with a different morphology. One of the challenges to relate the morphology and the internal model is the difference in the number of parameters that define them. We propose three neural network architectures for solving this problem, along with a case study to evaluate their performance, namely saccadic movements in a robotic head. We generate a huge dataset to test the performance of the proposed architectures. Our results suggest that the best solution is provided by the parallel neural network, due to the fact that the trained weights are independent of one another.

Visual motion estimation is challenging, due to high data rates, fast camera motions, featureless or repetitive environments, uneven lighting, and many other issues. In this work, we propose a two-layer approach for visual odometry with stereo cameras, which runs in real-time and combines feature-based matching with semi-dense direct image alignment. Our method initializes semi-dense depth estimation, which is computationally expensive, from motion that is tracked by a fast but robust feature point-based method. By that, we are not only able to efficiently estimate the pose of the camera with a high frame rate, but also to reconstruct the 3D structure of the environment at image gradients, which is useful, e.g., for mapping and obstacle avoidance. Experiments on datasets captured by a micro aerial vehicle (MAV) show that our approach is faster than state-of-the-art methods without losing accuracy. Moreover, our combined approach achieves promising results on the KITTI dataset, which is very challenging for direct methods, because of the low frame rate in conjunction with fast motion.

This paper describes a view-based outdoor navigation method. Navigation in outdoor can be divided into two levels; the global level deals with localization and subgoal selection, while the local level deals with safe navigation in a local area. We adopt an improved version of SeqSLAM method for global-level localization, which can cope with changes of robot’s heading and speed as well as view changes using very wide-angle images and a Markov localization scheme. The global level provides the direction to move and the local level repeatedly sets subgoals with local mapping using 3D range sensors. We implemented these global and local level methods on a mobile robot and conducted on-line navigation experiments.

In this paper, the integration of OpenStreetMap (OSM) geodata to a robot system which focuses on autonomous off-highway driving is presented. It is shown, how the OSM data is enriched with other data sources and how the map information is processed to generate a path that fits to the capabilities of the robot. Based on the map information, the quality of Global Satellite Navigation System (GNSS) signals is estimated and incorporated into the routing process, e.g. to avoid path with a high probably of GNSS disturbances. Furthermore, it is demonstrated how the robot’s localization based on a Carlson filter can be improved by these estimations.

This paper presents a lane change detection and filtering approach for precise longitudinal localization. Maps, the road which is traveled along and the trajectory of the vehicle are used as the only inputs. Straight-road transformation is proposed for lane change detection on the curve, and the filtering algorithm is designed for online positioning process. Experiments demonstrate the improvement of longitudinal position precision by lane change detection and filtering, and show the application of this approach on terrain localization.

Robots perceive their environment by processing continuous streams of data, which can be very naturally modelled as a dataflow graph. The development of new perception algorithms is often an iterative process, involving the investigation of a set of parameters and their influence on the system. The amount of immediate feedback available to the developer can make these influences more obvious and can therefore speed up development. We present a framework based on synchronous dataflow and event-based message passing that forms the basis of a visual programming language for rapid prototyping of robotic perception systems. We explicitly model algorithmic parameters in the dataflow graph, which results in a more expressive feature set. We provide an open-source implementation, consisting of a user interface for immediate feedback and interactive manipulation of dataflow algorithms and an independent execution framework that can be directly used on any robot.

In this work, we propose to exploit depth information to build a pose-invariant face recognition algorithm from RGB-D data. Our approach first estimates the head pose and then generates a frontal view for those faces that are rotated with respect to the frame of the camera. Then, some interest points of the face are detected by means of a Random Forest applied to the RGB image and they are used as keypoints where to compute feature descriptors. Around these points and their 3D counterpart, we extract both 2D and 3D local descriptors, which are then concatenated and classified by means of a Support Vector Machine trained in “one-versus-all” fashion. In order to validate the accuracy of the system with data from consumer RGB-D sensors, we created the IAS-Lab RGB-D Face Dataset, a new public dataset in which RGB-D data are acquired with a second generation Microsoft Kinect. The reported experiments show that the depth-aided approach we propose allows to improve the recognition rate up to 50 %.

In this paper, we investigate a model-free method for robot programming referred to as view-based teaching/playback. It uses neural networks to map factor scores of input images onto robot motions. The method can achieve greater robustness to changes in the task conditions, including the initial pose of the object, as compared to conventional teaching/playback. We devised an online algorithm for adaptively switching between range and grayscale images used in view-based teaching/playback. In its application to pushing tasks using an industrial manipulator, view-based teaching/playback using the proposed algorithm succeeded even under changing lighting conditions. We also devised an algorithm to cope with occlusions using subimages, which worked successfully in experiments.

In this research, we aimed at designing a portable and safe dual arm robot to help people, especially for the elder or disable people in their daily life. With the consideration of portable, we limited the weight of the robot to be lower than 7 kg. The weight enables a person to lift the robot, even though by single arm. To overcome this challenging, most of the parts of the robot was designed to be manufactured by plastic. On the other hand, for the safety, we design a new type of passive compliant unit to sense the torque of the joint and buffer the impact of collisions. In each arm of the robot, there were six degrees of freedom, three in the shoulder and one in the elbow. In order to detect and buffer the impact of collision which occurred in any position of the arm, passive compliant unit was installed in both shoulder and elbow. In addition, there was also two degree of freedoms in the wrist joints for the robot to adjust the orientation of the end effector. In order to examine our design, the whole robot was manufactured by 3D printer with ABS material, except the motor, bearing and the screw. Finally, an experiment was conducted to test the proposed dual arm robot’s basic performance. The result showed that the payload of the robot was up to 500 g and the maximum reach is up to 400 mm. In addition, utilized the passive compliant units of the shoulder and elbow, the robot arm was able the buffering impact.

A device for measuring the inertia parameters of rigid bodies has been presented in this paper. It’s actually a 3-URU pure rotation parallel mechanism. To improve the measuring accuracy, an adjusting mechanism composed of dovetail guides, bevel gears and a motor in the measuring device, is adopted to facilitate the adjustments of the center of gravity of the rigid body. Only three 16-bits encoders and three load cells are needed. The direct kinematic model of the parallel mechanism is built and the kinematic analysis is accomplished. The dynamic behaviors are investigated by Adams modeling. Simulation results are presented and show that the design of the device can meet the technical demand on inertia parameter measurement.

Restoring the hand functionality of partial amputees requires a myoelectric prosthetic hand, which is a robotic hand controlled by myoelectric signals from the skin surface and has the potential to restore human hand functionality. Most myoelectric hands have been developed for forearm amputees, while those for partial-hand amputees are few in number despite the higher numbers of the latter. Partial-hand amputees have limited hand functionality and cannot grasp and manipulate various objects in a manner comparable to individuals with a healthy hand; thus, they require a myoelectric hand. In this study, design issues were identified, and three-dimensional computer-aided design was used to propose an integrated skeleton and housing with a supporting socket. A passive thumb mechanism with motion in the remaining part of the hand was developed where the motion control system is based on the amputee’s muscles. An amputation system is proposed comprising mixed metacarpal and center-part cuts. A prototype was constructed, and its gripping functionality was evaluated. The results demonstrated an enhanced gripping performance compared to the non-use of prosthetics, which attests to the viability and effectiveness of the system.

The tele-operated in-vessel inspection system is required for maintenance in nuclear fusion environment. A Multi-joint Foldable Robot (MFR) is designed for Experimental Advanced Superconducting Tokamak (EAST). The whole MFR system has been designed and implemented. The motion control experiments show that the design is reasonable and effective. The MFR system developed can be used to perform inspection task in a nuclear environment.

The article describes the remote cutting tools designed for pipe cutting at Tokamak. Because of the special working environment, there are some design requirements for the cutting tools: the cutting swarf should be removed effectively, the quality of cut necessary for re-welding, and the cutting tools should be compact to fit into the maintenance manipulator for remote handling. The cutting tool consists of two parts: the cutting part and the sealing part. The two parts of cutting tool are designed to cut pipes and retain all cutting debris, respectively. A remote computer is used to input the commands and display data. During the cutting operation, the programmable controller has been used for controlling the cutting tool. The commands through the controller to motor drivers, implement the desired action.

Trajectory planning for manipulators with flexible links is a complicated task that plays an important role in design and application of manipulators. This paper is concerned with optimal trajectory planning for a two-link manipulator consisting of a macro flexible curved link and a micro rigid link for a point-to-point motion task. Absolute nodal coordinate formulation (ANCF) is used to derive the dynamic equations of the flexible curved link, an optimal trajectory method is adopted to generate the trajectory that minimizes the vibration of the flexible curved link. The Hamiltonian function is formed and the necessary conditions for optimality are derived from the Pontryagin’s minimum principle (PMP). The obtained equations form a two-point boundary value problem (TPBVP) which can be solved by numerical techniques. Finally, simulations for the two-link manipulator are carried out to demonstrate the efficiency of the presented method. The results illustrate the validity of the method to overcome the high nonlinearity nature of the whole system.

In this paper, the issue that traditional virtual reality (VR) system can’t be used in unstructured environment is addressed, and a novel predictive display method based on 3D scene reconstruction online is proposed. In this method, the virtual environment is modeled and reconstructed online by the structured light, so the consistency with the real world can be ensured; An adaptive random sample consensus (ARANSAC) algorithm is proposed to denoise the point cloud when modeling the environment, which adjusts the parameters of the standard RANSAC algorithm adaptively; A rigid contact model based haptic rendering algorithm is adopted to generate the force feedback directly from the manipulator’s dynamics, which eliminates the time delay and security problem introduced by the feedback force and increases the frequency of force feedback; Moreover, a low frequency filter is designed to keep the virtual force smooth. Finally, the effectiveness of the proposed method under unstructured environment is demonstrated by experiments.

In nuclear fusion research, inspecting tokamak vessel by controlling remote-handling robot is promising. To achieve collision-free, precise robot movement in tokamak vessel, it is necessary for the control system to perform motion planning, closed-loop control, interface operation and other functions. In this paper, a control system based on Ethernet and CAN network for a developed snake-like robot for inspecting tokamak vessel is proposed. Then, the robot kinematics model is built. Motion planning procedure is completed by a stepwise iterative algorithm. And a dual-loop control method for each joint is realized. A control software is developed to perform control computation and interface operation. Finally, the effectiveness of the control system for the robot is verified by actual experiments.

The tele-operated transfer cask is required for maintenance in nuclear fusion environment. A type of the transfer cask system, namely TECA (Tokamak Equipment CAsk) is designed for Experimental Advanced Superconducting Tokamak (EAST). The whole TECA system schemes, composed of the Air Cushion Vehicle (ACV), the Docking Pallet (DP), and the Conservation Cask (CC), are designed, and the prototype is accomplished. The motion and control of the ACV, the DP, and the DSD are detailedly analyzed. The motion control experiments validate the TECA applicable.

As the intra-scene of Tokamak chamber contains many repetitive textures, the traditional 3D reconstruction method based on the feature descriptor would be difficult to work well since the image matching algorithms based on feature descriptor are unstable and may fail sometimes in this environment. To address this problem, a novel multilevel matching algorithm is proposed, which uses the structural characteristics of Tokamak chamber as prior knowledge to find reliable correspondence points between two images. Firstly, each image is divided into basic structure regions. Then, to obtain the corresponding relation of structure regions from multiple images, we take the preliminary matching on the structural framework. The feature points is matched inner the structure regions to ensure the correctness of the feature matching. To testify the effectiveness of the proposed algorithm, it is applied to repetitive texture images captured in the Tokamak chamber, and the experimental results show that more correct matching points are acquired, smooth and clear 3D point-cloud data are generated, and high accurate and integrated reconstruct results are obtained.

Maintenance for nuclear fusion vessel is crucial, yet it faces great difficulty due to the complex internal physical and geometric conditions. For this reason, such remote control mobile robot system is required to do the inboard detection and maintenance work instead of human beings. Under the above background, this paper presents a wormlike creeping robot system working on the V-shaped circular slot at the bottom of the EAST nuclear fusion vessel, and constructs the engineering prototype. The wormlike creeping robot adopts three-part structure, which consists of fore segment, mid segment and back segment connected by bidirectional universal joint. The fore and back segments stretch the paws to contact the surface of V-shaped slot, while the mid segment realizes the movement of robot by axial expansion. Functional analysis is devoted to modules of robot system. For the requirements of walking stability on the V-shaped slot, creeping gait planning is analyzed. Finally, the engineering prototype is running in the simulated vessel to test the synthesis movement properties and evaluate the mechanical properties. The test shows that the wormlike creeping robot system has good feasibility and effectiveness.

An importance of accurate position estimation in the field of mobile robot navigation cannot be overemphasized. In case of an outdoor environment, a global positioning system (GPS) is widely used to measure the position of moving objects. However, the satellite based GPS does not work indoors. In this paper, we propose a novel indoor positioning system (IPS) that uses calibrated camera sensors and 3D map information. The IPS information is obtained by generating a bird’s-eye image from multiple camera images; thus, our proposed IPS can provide accurate position information when the moving object is detected from multiple camera views. We evaluate the proposed IPS in a real environment in a wireless camera sensor network. The results demonstrate that the proposed IPS based on the camera sensor network can provide accurate position information of moving objects.

Autonomous driving technology has become a spotlight in recent years. Of all the factors related to autonomous driving, safety should be first considered. A safe global trajectory should be planned at beginning and local safe trajectory should be planned according to the situations in real time. Due to this, the intelligent vehicles must know where they are in real time to do the next control steps. In this paper, a high-precision localization framework for intelligent vehicles is proposed. A vertical low-cost LIDAR is used for mapping and live data collection. High-precision maps are generated by projecting laser scans along the survey trajectory produced by trajectory filter. When localizing, an improved matching method particle Iterative Closet Point is proposed. Using this particle ICP, not only the matching precision is improved, but also the computing time decreases remarkably, which helps to make the algorithm real-time. Decimeter-level precision can be achieved by the validation of experiments. The results show much benefit for safe driving by this Monte Carlo framework.

The robots equipped with wireless networking modules can act as mobile routers to bridge the communications of a network. How to adjust the robot network topology and the motion of mobile routers adaptively to support the communication connectivity to mobile users in group task execution is still remained challenging. In this paper, we addressed this challenging by developing a unified motion controller to drive the robots to approach their individual task region while maintaining the desired network topology and keeping collision-free with obstacles in environment. To achieve this, a new concept termed rubber communication model is first proposed to evaluate the real communication signal, which enables adding and removing communication links amongst robots. Then, a continuous model for collision avoidance is utilized for avoiding obstacles. Together with the rubber communication model and continuous model for collision avoidance, the tasks assigned to the robots are modeled as series of geometrical task regions which is formulated with the regional reaching constraint function. The three models are utilized in building the potential field function, based on which a bounded control input is generated for multirobot control. Simulations are finally performed on a group of mobile robots to demonstrate the effectiveness of the proposed controller.

Intelligent wearable device is currently a hot spot, while the majority of smart wearable devices are concerned about the health of human information monitoring. In the study of the elderly wearable devices, the task is to find the elderly who accidentally fall, and give warning to others immediately. In order to effectively recognize the falling posture, this paper proposes a falling motion recognition algorithm combined with Kinect and wearable sensors. Although acceleration amplitude detected by three-axis accelerometer inside the wearable device can also recognize the falling motion, it is not accurate enough to distinguish the weightlessness and people’s posture after falling. Therefore, Kinect’s skeleton data is used to construct the feature vector of falling posture, and Support Vector Machines (SVM) is used to classify it. The experiments show that the accuracy of falling recognition is over 98 %, the real-time performance has been greatly improved as well.

Task modeling and task planning are very important in robotic systems especially for large-scale nondeterministic problems. Two widely studied models (the classical planning model and the Markov Decision Process (MDP) model) are inapplicable to such problems due to either inherently assumed determinism or dimensional explosion. An amalgamation of these two results in a new model which is proposed in this study under the name “Reduced Markov Decision Process” (RMDP) model. This new model simplifies the conventional MDP model by reducing the branching factor of state transitions. Further, based on the RMDP model, a modified Dynamic Programming (DP) algorithm is proposed. The RMDP model also facilitates online learning that adapts the model to environmental changes. A “forgetting” model is employed for this online adjustment. In the experiment, a ubiquitous robotic system is implemented for robotic bar-tending task. The results demonstrate that the model conveniently facilitates online-updating to better match the real environment.

In this work, we address the problem of human skeleton estimation when multiple depth cameras are available. We propose a system that takes advantage of the knowledge of the camera poses to create a collaborative virtual depth image of the person in the scene which consists of points from all the cameras and that represents the person in a frontal pose. This depth image is fed as input to the open-source body part detector in the Point Cloud Library. A further contribution of this work is the improvement of this detector obtained by introducing two new components: as a pre-processing, a people detector is applied to remove the background from the depth map before estimating the skeleton, while an alpha-beta tracking is added as a post-processing step for filtering the obtained joint positions over time. The overall system has been proven to effectively improve the skeleton estimation on two sequences of people in different poses acquired from two first-generation Microsoft Kinect.