Table Cleaning through Task and Motion Planning with Force Control

William Shen, MIT - 6.843 Manipulation 2021 Final Project

YouTube Presentation: https://www.youtube.com/watch?v=j5t2DlDHMoU

Paper (PDF)

We define four primitive behaviors: (1) pick, (2) place, (3) wipe and (4) drag using a low-level Cartesian impedance controller and RRT-based motion planner.

We use Task and Motion Planning to build a system which can reason and solve long-horizon problems by composing these primitives.

To ensure the project is feasible within the limited timeframe, we make several assumptions:

Please direct questions to willshen at mit.edu

PDDL

Domain PDDL  |  Problem PDDL for Scenario in Videos

Overall Demo

Static View (7x)

Tracking View (5x)

Our implementation of collision checking when holding an object is still buggy.


Force Control

We implemented an open-loop Cartesian Impedance Controller.

To exert forces into the table, we can define setpoints underneath the table.

Green arrows depict contact forces.

Dragging (2x)

Wiping (2x)


Other Examples (updated 2021-12-14)

Pick and Place with Wiping (4x)

Wiping multiply objects (6x)

Based on an earlier implementation - note the RRT dance at the end when placing the sponge.

We did not rerun RRT or continuously grow the tree here.

Note order of wiping of cubes was pre-defined in this example.

Earlier YouTube Videos for Wiping and Dragging