|R obotic A lgorithms & M otion P lanning L aboratory|
A Geometric Reasoning and Motion Planning ToolkitThe main objective of this research is to develop and implement advanced and sophisticated motion planning and geometric reasoning tools for manipulator robots. Some examples are automatic collision detection, collision-free motion planning of paths for single and multiple robot arms, automatic grasping with simple end-effectors, following a desired end-effector path, particularly for redundant robots. These have been implemented into a general purpose, well packaged and stable software -- a geometric reasoning and motion planning toolkit, called MPK (Motion Planning Toolkit). MPK facilitates implementation and testing of motion planners and robotic algorithms with a geometric flavor.
The library is available upon request to Kamal Gupta.
Sensor-Based Motion PlanningA main objective here is to develop formal and practical approaches to incorporate sensing into geometric reasoning, and develop a general framework for sensor-based path planning for general robot-sensor systems. While most sensor-based planning research has focussed on mobile robots, our emphasis is to develop and implement an approach that is practical and efficient for more complex robots such as eye-in-hand systems, mobile-manipulators, etc.
A particular recent emphasis has been on the development of the SFU Eye-in-Hand system -- a PUMA 560 manipulator arm with a wrist mounted area scan laser range finder. The sensor-based planner employs SBIC-PRM (Sensor based incremental construction of probabilistic roadmap) and an MER (Maximum Entropy Reduction) criterion based view planning algorithm. The robot is started in an unknown and cluttered environment, and typically the planner is able to reach its goal configuration, planning as it senses, and avoiding collisions with the obstacles (unknown to it in the beginning) in about twenty minutes.
The same approach has also been extended to manipulators with "skin" sensors -- an array of proximity/contact sensors distributed around the whole manipulator.
Model-Based Motion Planning for Systems with Many DOFsIn the past, we have established a formal deterministic framework (Sequential Framework) for developing practical motion planners for many DOFs manipulator arms. The resulting algorithm has been applied to develop practical motion planners for many DOFs manipulators in industry and at other research institutions. The technology was transferred to ISE Ltd., a Vancouver company.
A complementary approach has been developed in collaboration with researchers from INRIA Rhone-Alpes in Grenoble, France and from Universidad de las Americas, Puebla, Mexico. The same framework has been applied to develop a novel approach to solve inverse kinematics problems for redundant robots.
Efficient Representations for Collision DetectionA key aspect of most motion planners is efficient collision detection, which in turn depends on the geometric representation used. For sensor-based planning, where CAD models are not available (environment is not known and must be sensed), volumetric representations of environment are much easier to obtain. This research deals with various ways to augment such representations so that collision detection is efficient.
Dextrous Manipulation PlanningManipulation by artificial multi-fingered hands has received a lot of attention, but mostly in mechanics and control aspect. Global motion planning for such complex systems has not been addressed in any significant way. It is a complex and fascinating problem from motion planning perspective since several constraints at different levels -- mechanics, kinematics, contact modeling, etc., need to be addressed. We have developed implemented planners that can automatically plan manipulation motions of a multi-fingered hand to re-configure objects held by the hand.
Part OrientationPart orientation is a basic step in most automatic assembly processes. However, design of such devices, in practice, is still mostly a black art governed mainly by the intuition of the designer. Our aim is to use basic mechanics and automate, at least partly, this design process and furthermore, design the part orienting devices so that it can be re-configured easily. Our approach spans several directions: incorporating sensors, design of algorithms and planners to automatically re-configure the device, etc.
© 2010 Simon Fraser University - Engineering Science - Robotic Algorithms and Motion Planning Laboratory. All Rights Reserved