Towards Autonomous Robotic Systems

Table of Contents

1. Frontmatter

2. Object Manipulation and Locomotion

   1. Frontmatter

   2. Trajectory Optimization for High-Power Robots with Motor Temperature Constraints

      Wei Xin Tan, Martim Brandão, Kenji Hashimoto, Atsuo Takanishi

      Abstract

      Modeling heat transfer is an important problem in high-power electrical robots as the increase of motor temperature leads to both lower energy efficiency and the risk of motor damage. Power consumption itself is a strong restriction in these robots especially for battery-powered robots such as those used in disaster-response. In this paper, we propose to reduce power consumption and temperature for robots with high-power DC actuators without cooling systems only through motion planning. We first propose a parametric thermal model for brushless DC motors which accounts for the relationship between internal and external temperature and motor thermal resistances. Then, we introduce temperature variables and a thermal model constraint on a trajectory optimization problem which allows for power consumption minimization or the enforcing of temperature bounds during motion planning. We show that the approach leads to qualitatively different motion compared to typical cost function choices, as well as energy consumption gains of up to 40%.

   3. SPGS: A New Method for Autonomous 3D Reconstruction of Unknown Objects by an Industrial Robot

      Cihan Uyanik, Sezgin Secil, Metin Ozkan, Helin Dutagaci, Kaya Turgut, Osman Parlaktuna

      Abstract

      This paper presents the first findings of a new method called surface profile guided scan (SPGS) for 3D surface reconstruction of unknown small-scale objects. This method employs a laser profile sensor mounted on an industrial manipulator, a rotary stage, and a camera. The system requires no prior knowledge on the geometry of the object. The only information available is that the object is located on the rotary table, and is within the field of view of the camera and the working space of the industrial robot. First a number of surface profiles in the vertical direction around the object are generated from captured images. Then, a motion planning step is performed to position the laser sensor directed towards the profile normal. Finally, the 3D surface model is completed by hole detection and scanning process. The quality of surface models obtained from real objects with our system prove the effectiveness and the versatility of our 3D reconstruction method.

   4. A Modified Computed Torque Control Approach for a Master-Slave Robot Manipulator System

      Ololade O. Obadina, Mohamed Thaha, Kaspar Althoefer, M. Hasan Shaheed

      Abstract

      A modified computed torque controller, adapted from the standard computed torque control law, is presented in this paper. The proposed approach is demonstrated on a 4-degree of freedom (DOF) master-slave robot manipulator and the modified computed torque controller gain parameters are optimized using both particle swarm optimization (PSO) and grey-wolf optimization algorithms. The feasibility of the proposed controller is tested experimentally and compared with its standard computed torque control counterpart. Controller tuning/optimization is carried out offline in the MATLAB/Simulink environment, and results show that the proposed controller is feasible, and performs impressively.

   5. Data Synthesization for Classification in Autonomous Robotic Grasping System Using ‘Catalogue’-Style Images

      Michael Cheah, Josie Hughes, Fumiya Iida

      Abstract

      The classification and grasping of randomly placed objects where only a limited number of training images are available, remains a challenging problem. Approaches such as data synthesis have been used to synthetically create larger training data sets from a small set of training data and can be used to improve performance. This paper examines how limited product images for ‘off the shelf’ items can be used to generate a synthetic data set that is used to train a that allows classification of the item, segmentation and grasping. Experiments investigating the effects of data synthesis are presented and the subsequent trained network implemented in a robotic system to perform grasping of objects.

   6. BounceBot: A One-Legged Jumping Robot

      James Rogers, Katherine Page-Bailey, Ryan Smith

      Abstract

      This paper describes the design and development of a jumping robot made from readily available components and 3D printed parts. This robot is designed to traverse obstacles that are too large for conventional locomotion methods, utilising elastic potential energy to store and release kinetic energy at differing rates. Rapidly releasing built up energy in this manner enables a small light-weight actuator to exceed its continuous torque output. This is used to accelerate the robot vertically and jump over an obstacle up to ten times its own height. Use of soft 3D printed materials allow for the robot to resist the impact caused by landing onto/jumping into obstacles. Due to its performance and availability/cost of its parts, this prototype provides a good platform for further research into this viable yet under-developed locomotion method. As the design is open source, researchers are free to use the details contained in this report along with the documentation available online. The concept can be used in a range of situations involving locomotion over uneven terrain. Potential projects include hazardous disaster site evaluation, planet exploration, and search and rescue.

   7. Estimating Grasping Patterns from Images Using Finetuned Convolutional Neural Networks

      Ashraf Zia, Bernard Tiddeman, Patricia Shaw

      Abstract

      Identification of suitable grasping pattern for numerous objects is a challenging computer vision task. It plays a vital role in robotics where a robotic hand is used to grasp different objects. Most of the work done in the area is based on 3D robotic grippers. An ample amount of work could also be found on humanoid robotic hands. However, there is negligible work on estimating grasping patterns from 2D images of various objects. In this paper, we propose a novel method to learn grasping patterns from images and data recorded from a dataglove, provided by the TUB Dataset. Our network retrains, a pre-trained deep Convolutional Neural Network (CNN) known as AlexNet, to learn deep features from images that correspond to human grasps. The results show that there are some interesting grasping patterns which are learned. In addition, we use two methods, Support Vector Machines (SVM) and hotelling’s T² test to demonstrate that the dataset does include distinctive grasps for different objects. The results show promising grasping patterns that resembles actual human grasps.

3. Soft and Bioinspired Robotics

   1. Frontmatter

   2. Easy Undressing with Soft Robotics

      Tim Helps, Majid Taghavi, Sarah Manns, Ailie J. Turton, Jonathan Rossiter

      Abstract

      Dexterity impairments affect many people worldwide, limiting their ability to easily perform daily tasks and to be independent. Difficulty getting dressed and undressed is commonly reported. Some research has been performed on robot-assisted dressing, where an external device helps the user put on and take off clothes. However, no wearable robotic technology or robotic assistive clothing has yet been proposed that actively helps the user dress. In this article, we introduce the concept of Smart Adaptive Clothing, which uses Soft Robotic technology to assist the user in dressing and undressing. We discuss how Soft Robotic technologies can be applied to Smart Adaptive Clothing and present a proof of concept study of a Pneumatic Smart Adaptive Belt. The belt weighs only 68 g, can expand by up to 14% in less than 6 s, and is demonstrated aiding undressing on a mannequin, achieving an extremely low undressing time of 1.7 s.

   3. Biomimetic Knee Design to Improve Joint Torque and Life for Bipedal Robotics

      Alexander G. Steele, Alexander Hunt, Appolinaire C. Etoundi

      Abstract

      This paper details the design, construction, and performance analysis of a biologically inspired knee joint for use in bipedal robotics. The design copies the condylar surfaces of the distal end of the femur and utilizes the same crossed four-bar linkage design the human knee uses. The joint includes a changing center of rotation, a screw-home mechanism, and patella; these are characteristics of the knee that are desirable to copy for bipedal robotics. The design was calculated to have an average sliding to rolling ratio of 0.079, a maximum moment arm of 2.7 in and a range of motion of 151°. This should reduce wear and perform similar to the human knee. Prototypes of the joint have been created to test these predicted properties.

   4. Evaluating the Radiation Tolerance of a Robotic Finger

      Richard French, Alice Cryer, Gabriel Kapellmann-Zafra, Hector Marin-Reyes

      Abstract

      In 2024, The Large Hadron Collider (LHC) at CERN will be upgraded to increase its luminosity by a factor of 10 (HL-LHC). The ATLAS inner detector (ITk) will be upgraded at the same time. It has suffered the most radiation damage, as it is the section closest to the beamline, and the particle collisions. Due to the risk of excessive radiation doses, human intervention to decommission the inner detector will be restricted. Robotic systems are being developed to carry out the decommissioning and limit radiation exposure to personnel. In this paper, we present a study of the radiation tolerance of a robotic finger assessed in the Birmingham Cyclotron facility. The finger was part of the Shadow Grasper from Shadow Robot Company, which uses a set of Maxon DC motors.

   5. Soft Pneumatic Prosthetic Hand

      Jan Fras, Kaspar Althoefer

      Abstract

      Conventional prosthetic devices are heavy, expensive and rigid. They are complex, fragile and require sophisticated control strategies in order to deal with the grasping and manipulation tasks. In this paper we propose a new pneumatic soft prosthetic hand that is very simple to control due to its compliant structure and cheap in production. It is designed to be easily reshaped and resized to adapt easily to each individual user preferences. It is designed to be frequently changed whenever a child patient require a bigger size or whenever the old one is worn out or broken. Since it is soft and compliant it can be safely used even by small children without a risk of harmful mechanical interaction.

4. Path Planning and Autonomous Vehicles

   1. Frontmatter

   2. Tabu Temporal Difference Learning for Robot Path Planning in Uncertain Environments

      Changyun Wei, Fusheng Ni

      Abstract

      This paper addresses the robot path planning problem in uncertain environments, where the robot has to avoid potential collisions with other agents or obstacles, as well as rectify actuation errors caused by environmental disturbances. This problem is motivated by many practical applications, such as ocean exploration by underwater vehicles, and package transportation in a warehouse by mobile robots. The novel feature of this paper is that we propose a Tabu methodology consisting of an Adaptive Action Selection Rule and a Tabu Action Elimination Strategy to improve the classic Temporal Difference (TD) learning approach. Furthermore, two classic TD learning algorithms (i.e., Q-learning and SASRA) are revised by the proposed Tabu methodology for optimizing learning performance. We use a simulated environment to evaluate the proposed algorithms. The results show that the proposed approach can provide an effective solution for generating collision-free and safety paths for robots in uncertain environments.

   3. Modelling and Predicting Rhythmic Flow Patterns in Dynamic Environments

      Sergi Molina, Grzegorz Cielniak, Tomáš Krajník, Tom Duckett

      Abstract

      We present a time-dependent probabilistic map able to model and predict flow patterns of people in indoor environments. The proposed representation models the likelihood of motion direction on a grid-based map by a set of harmonic functions, which efficiently capture long-term (minutes to weeks) variations of crowd movements over time. The evaluation, performed on data from two real environments, shows that the proposed model enables prediction of human movement patterns in the future. Potential applications include human-aware motion planning, improving the efficiency and safety of robot navigation.

   4. Extending Deep Neural Network Trail Navigation for Unmanned Aerial Vehicle Operation Within the Forest Canopy

      Bruna G. Maciel-Pearson, Patrice Carbonneau, Toby P. Breckon

      Abstract

      Autonomous flight within a forest canopy represents a key challenge for generalised scene understanding on-board a future Unmanned Aerial Vehicle (UAV) platforms. Here we present an approach for automatic trail navigation within such an unstructured environment that successfully generalises across differing image resolutions - allowing UAV with varying sensor payload capabilities to operate equally in such challenging environmental conditions. Specifically, this work presents an optimised deep neural network architecture, capable of state-of-the-art performance across varying resolution aerial UAV imagery, that improves forest trail detection for UAV guidance even when using significantly low resolution images that are representative of low-cost search and rescue capable UAV platforms.

   5. Virtual Environment for Training Autonomous Vehicles

      Jerome Leudet, Tommi Mikkonen, François Christophe, Tomi Männistö

      Abstract

      Driver assistance and semi-autonomous features are regularly added to commercial vehicles with two key stakes: collecting data for training self-driving algorithms, and using these vehicles as testbeds for these algorithms. Due to the nature of algorithms used in autonomous vehicles, their behavior in unknown situation is not fully predictable. This calls for extensive testing. In this paper, we propose to use a virtual environment for both testing algorithms for autonomous vehicles and acquiring simulated data for their training. The benefit of this environment is to able to train algorithms with realistic simulated sensor data before their deployment in real life. To this end, the proposed virtual environment has the capacity to generate similar data than real sensors (e.g. cameras, LiDar, ...). After reviewing state-of-the-art techniques and datasets available for the automotive industry, we identify that dynamic data generated on-demand is needed to improve the current results in training autonomous vehicles. Our proposition describes the benefits a virtual environment brings in improving the development, quality and confidence in the algorithms.

   6. Comparing Model-Based and Data-Driven Controllers for an Autonomous Vehicle Task

      Erwin Jose Lopez Pulgarin, Tugrul Irmak, Joel Variath Paul, Arisara Meekul, Guido Herrmann, Ute Leonards

      Abstract

      The advent of autonomous vehicles comes with many questions from an ethical and technological point of view. The need for high performing controllers, which show transparency and predictability is crucial to generate trust in such systems. Popular data-driven, black box-like approaches such as deep learning and reinforcement learning are used more and more in robotics due to their ability to process large amounts of information, with outstanding performance, but raising concerns about their transparency and predictability. Model-based control approaches are still a reliable and predictable alternative, used extensively in industry but with restrictions of their own. Which of these approaches is preferable is difficult to assess as they are rarely directly compared with each other for the same task, especially for autonomous vehicles. Here we compare two popular approaches for control synthesis, model-based control i.e. Model Predictive Controller (MPC), and data-driven control i.e. Reinforcement Learning (RL) for a lane keeping task with speed limit for an autonomous vehicle; controllers were to take control after a human driver had departed lanes or gone above the speed limit. We report the differences between both control approaches from analysis, architecture, synthesis, tuning and deployment and compare performance, taking overall benefits and difficulties of each control approach into account.

   7. An Improved Robot Path Planning Model Using Cellular Automata

      Luiz G. A. Martins, Rafael da P. Cândido, Mauricio C. Escarpinati, Patricia A. Vargas, Gina M. B. de Oliveira

      Abstract

      Bio-inspired techniques have been successfully applied to the path-planning problem. Amongst those techniques, Cellular Automata (CA) have been seen a potential alternative due to its decentralized structure and low computational cost. In this work, an improved CA model is implemented and evaluated both in simulation and real environments using the e-puck robot. The objective was to construct a collision-free path plan from the robot initial position to the target position by applying the refined CA model and environment pre-processed images captured during its navigation. The simulations and real experiments show promising results on the model performance for a single robot.