6.881 Project Ryan Sandzimier and Filippos Sotiropoulos
In this project we implement task and motion planning for an object rearrangement problem. A collection of cube blocks are arranged in a densely packed configuration by utilising a combination of pick-and-place as well as non-prehensile pushing actions. Utilising a Probabilistic Roadmap planner to choose control actions to relocate single blocks and a recursive algorithm, called Piecewise Linear Non-Monotone Rearrangement Search [1], to plan the sequence of blocks to move, we are able to successfully plan a sequence of actions to rearrange handfuls of blocks from a random original configuration to the densely packed grid. The method is demonstrated on an experimental setup.
[1] A. Krontiris and K. E. Bekris, “Dealing with difficult instances of object rearrangement,” in Robotics: Science and Systems (RSS) , Rome, Italy, 07/2015 2015.
A video demonstrating our implementation can be seen here: https://youtu.be/CdenErVEkS4