Graphics and Robotics

This paper addresses some fundamental topics which have to be regarded during the design, implementation and usage of graphical robot Simulation systems.

After a short summary of robot Simulation system purposes, we derive the basic requirements, which have to be satisfied, in order to obtain a useful Simulation system. We show, that different modeling steps are necessary to build up an internal representation of a robot arm and the robot work cell, which can be used to simulate and visualize the execution of robot tasks. The main parts of such a model are the geometrical and the kinematical representation. Different geometrical representations and possible modeling techniques are presented. In addition, general considerations about the computation of the robot kinematics and its inverse, which plays an important role in off-line robot program development, are made. An affixment technique which is used to combine the kinematical and the geometrical model is introduced and the working principle is demonstrated with the help of a modeled example world.

Finally we focus on some special aspects of robot Simulation such as the graphical visualization and the different interpretations of time within a Simulation system; in addition, we outline the main Software modules, which are more or less incorporated in each graphical robot Simulation system.

Many of the discussed principles of graphical robot simulation are illustrated by means of a PC-based graphical robot simulation system. This system was designed and implemented at our institute and is currently used to teach students some principles of robotics within practical courses.

Up to now robot task execution is limited to move the manipulator on the joint or cartesian level from one position into another one. We propose the so-called TeleSensorProgramming concept that uses sensory perception to bring local autonomy onto the manipulator level. This approach is applicable both in the real robot’s world and in the simulated one. Beside the graphical offline programming concept the range of application lies especially in the field of teleoperation with large time delays. The feasibility of graphically simulating the robot within its environment is extended by emulating different sensor functions like distance, force-torque and vision sensors to achieve a correct copy of the real system behaviour. These Simulation features are embedded into a task-oriented high-level robot programming approach. Sensor fusion aspects with respect to autonomous sensor controlled task execution are discussed as well as the interaction between the real and the simulated system.

In the field of raster graphics, numerous efficient algorithms — in Software as well as in hardware — have been developed in order to meet the requirements of interactivity. We investigate how to apply these algorithms to motion planning problems. It turns out that in particular the z-buffer algorithm is helpful for a rasterized implementation of the retraction approach known from literature, and for cell decomposition methods. The reason lies in the fact that calculation of Voronoi diagrams can be reduced to the calculation of visible surfaces.

This paper describes a complete method to detect collision among CSG modeled objects within a computer animation system. Since objects that are modeled with the constructive solid geometry paradigm can have very complex shapes, especially when they consist of curved primitives, collision detection is performed in three stages. Bounding volumes in each node of the CSG tree are used to determine whether a collision is likely to occur between each pair of objects. If overlapping bounds are discovered, spatial subdivision is used to reduce the complexity of the CSG objects for further analysis. In those voxels, which cover parts of both objects, a redundancy test is performed for each primitive to determine whether the objects interpenetrate. Curved primitives are adaptively approximated by circumscribed and inscribed polyhedrons.

Along with sophisticated techniques for natural visualization and rapidly increasing power of modern graphics Workstations, high-quality 3D graphics is becoming most attractive for design and Simulation. One area in which this new media proves especially useful is architecture and interior design. For example, the visualization of an office room or a building prior to its physical realization could help a designer to obtain realistic impressions of a construction while it is evolving and give free way to imagination at the same time. It is one of the aims that, eventually, a designer is able to explore, and interact with, a manipulable environment without wasting physical matter and with the ability to readily change the immaterial model.

After a short overview about different assembly planning techniques, our assembly planning system working with symbolic spatial relationships is presented. The spatial relations are used to define an assembly graph and to deduce automatically the homogeneous transformations between the parts of the assembly from symbolic information. The local depart space efficiently computed by symbolic relations delivers the basis for an estimation of a global translational depart vector. In addition, we have developed a concept of arbitrary assembly hierarchies which is integrated into the well-known cut-set method of the assembly graph and into the AND/OR-graph representation.

The use of robots for assembly tasks has not had the expected diffusion. As a matter of fact, assembling robots are today used only in specific environments like for example electronic board assembly or car mass-production. The main reason of this limited use is the robots high cost and the only apparent flexibility. Complete flexibility requires either evolution of the off-line programming systems which is still complex and machine dependent, and the realisation of efficient and reliable task level planning systems, directly related to both design and programming phases.

A product modelling approach is presented that supports the assembly planning task with feature models. Mating and connection relations between components of an assembly can be specified by adding connection features to a product model with a design-by-features approach. In a second feature model, handling features can be specified. From the feature models, a decomposition of the assembly into subassemblies, and all possible assembly sequences can be derived. The features can also support grasp planning and motion planning. The design-by-features approach is embedded in an interactive design system that supports multiple feature views.

Triangular B-splines are a new tool for the modeling of complex objects with non-rectangular topology. The new B-spline scheme is based on blending functions and control points and allows modeling piecewise polynomial surfaces of degree n that are C ^n-1 -continuous throughout [4, 15, 17]. An implementation of the new scheme at the University of Waterloo has succeeded in demonstrating the practical feasibility of the fundamental algorithms underlying the new scheme [6, 7].

Both in Graphics and Robotics one often faces the problem to construct “optimal” curves and surfaces subject to certain constraints. Here optimal typically means that the curve or surface minimizes length, area, curvature or other geometric properties.

To use the advantages of different surface representations, such as tensor-product surfaces, surfaces over triangular regions, implicitly defined surfaces, etc., it is necessary to integrate these surface types with a large variety of algorithms into one programming environment. Inheritance and polymorphism of object-oriented languages offer the opportunity to realize an implementation, which is not bound to a certain representation of the surface. The most reasonable way to benefit from such an object-oriented approach is to use existing algorithms and if necessary to develop new algorithms that are based on the functionality provided by as many surface representations as possible. An object-oriented design for surfaces is presented and the advantages of such a design are illustrated by examples.

We present a novel method for obtaining dense range maps, which is based on the combination of two (or more) colour cameras and the projection of a continuous colour pattern. The technique offers several advantages for use in robotics including the potential for very high speed of operation and for higher accuracy than achievable with other active triangulation techniques that employ discrete coloured light stripes or sequential binary-encoded monochrome patterns. The high speed makes the range image generation of fast moving objects in highly dynamic environments possible and thus enables a sensor based on this method to perform both coarse navigational tasks and the fine control of gripper Operations. In this approach a continuous colour pattern with uniform brightness is projected onto the objects in the scene.

Recently, the interactive exploration of volume data has received increasing attention in Scientific Visualization research. In this context one of the key problems is the machine perception of object surfaces prior to their three-dimensional rendering. We present an active vision approach exploiting interactivity features of modern visualization systems. By the integration of Computer Vision methods into the visualization pipeline immediate user control of the highly complex machine perception process becomes possible. On the basis of this concept we propose the vision camera, a tool for interactive vision during volume data walkthroughs. This camera model is characterized by a flexible front-plane which, under the control of user-specified Parameters and image features elastically matches to object surfaces, while shifted through a data volume. Thus, objects are interactively carved out and can be visualized by Standard volume visualization methods. Implementation and application of the model are described. Our results suggest that by the integration of human and machine vision new perspectives for data exploration are opened up.

We shall discuss a way to set up a reliable 3D office scene recognition and interpretation scheme for a home robot with the use of articicial vision only. It is a very difficult problem which has been studied by numerous scientists for many years. We do not pretend to describe the solution of this problem in the present paper, but simply to express our State of mind. Our efforts are concentrated on the manner to produce a well posed problem and on the use of an analysis/synthesis feedback loop in order to solve it. It is expected to produce a more robust solution than the visual open loop solutions already available in the literature.