Synergistic Task and Motion Planning With Reinforcement Learning-Based Non-Prehensile Actions

Robotic manipulation in cluttered environments requires synergistic planning among prehensile and non-prehensile actions. Previous works on sampling-based Task Motion Planning (TAMP) algorithms, e.g. PDDLStream, provide a fast generalizable solution for multi-modal manipulation. However, they are likely to fail scenarios where no collision-free grasping approaches can be sampled without preliminary manipulations. To extend the ability of we integrate vision-based Reinforcement Learning (RL) procedure, pusher . The pushing actions generated by eliminate interlocked situations make problem solvable. Also, algorithm evaluates providing rewards training process, thus learn avoid leading irreversible failures. proposed hybrid method is validated bin-picking implemented both simulation real world. Results show that effectively improve success ratio previous algorithm, while help skills.