Sampling-based planning for goal-directed robot control

Project Leaders: Robert Haschke, Michael Gienger
PhD Student: Matthias Behnisch

The Control Basis Framework of Grupen et al, composes complex robot movement from several basic task-oriented control units (e.g. collision avoidance, pre-shaping, grasping, reaching, walking, tracking, etc.) which can be flexibly combined in a parallel, sequential, or cascaded fashion [1,2].

While these reactive controllers may get stuck in local minima and eventually are not suitable for high-level task planning anymore, deliberative planning methods are able to find globally feasible solutions due to their increased preview horizon [3]. Nevertheless, they operate on a slower time scale and often suffer from a huge search space. Combining reactive controllers with deliberative planners in a hybrid approach, might fundamentally boost the performance of planning algorithms: While controllers can efficiently avoid e.g. local obstacles and find a path through a narrow passage, the deliberative planner can operate on a much coarser level of spatial resolution and thus save planning time.

Within this project we want to tackle research questions like: How to dynamically combine control units to compose complex action sequences? What are efficient mechanisms to describe such sequences and their transition behaviour? How the system can autonomously infer controller behaviour from observations and thus extend its control repertoire? How should we organise the interplay between a high-level deliberative and a low-level reactive planning and control?

Especially interesting is the application of this new planning paradigm to the situation of bimanual manipulation. Research questions arising within this context are: How should both hands, and gravitation as a third, be coordinated to achieve grasping an re-grasping of variously sized objects? For which objects are both hands needed, for which not?

[1] Brock, O., Fagg, A., Grupen, R., Platt, R., Rosenstein, M., and Sweeney, J. A Framework for Learning and Control in Intelligent Humanoid Robots, The International Journal of Humanoid Robotics, Volume 2, Number 3, September 2005.
[2] M. Gienger, H. Janssen, and C. Goerick, Taskoriented whole body motion for humanoid robots. In Proceedings of the IEEERAS/RSJ International Conference on Humanoid Robots, 2005.
[3] J. van den Berg, Path Planning in Dynamic Environments, PhD Thesis, Utrecht University, 2007