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:
We assume perception is solved and that we know the
precise pose of each object.
We use a human who is given a list of task plans to ‘filter
out’ unfeasible plans when we fail to motion plan at the
low-level.
We use manually determined grasps as our grasp sampler.
We only support straight line wiping in the y axis.