When You Robots Know What to Reach for - Contact-Based Motion Planning
Humans handle manipulation under uncertainty by touching objects to guide their motion and they use the feel of touch to react to any deviation. The key is to leverage the environment to reduce perception and motion uncertainty and to generate distinguishable feedback signals.
This is difficult for robots because representing the environment and dynamic interactions are either a weak approximation or too complex to plan robustly or efficiently. But we can design human-like manipulation strategies for robots and some motion planners can generate reactive robot behaviors, where the robot can take alternative actions based on senor event during execution. The challenge is to extend human-like manipulation strategies with robot-like reactiveness.
Description of Work
In our lab we implemented a sampling-based motion planner that combines controlled contact with configuration-space search (https://www.youtube.com/watch?v=mUbjuJ4pk8s). The student will extend this motion planner (CERRT) with knowledge about the environment’s geometry to improve planning speed. The planner and its solutions will be tested on our robots WAM and Panda.
For this project you need good C++ programming skills. Experience with the robotics-library (from our robotics and compbio classes) or is a plus.
Előd Páll 
Oliver Brock