A public implementation of TMIT* is available in its GitHub repository.
Task and Motion Informed Trees (TMIT*) is a practical approach to almost-surely asymptotically optimal integrated task and motion planning (TMP). It uses relaxed symbolic planning and anytime satisfaction of action preconditions to efficiently discover candidate symbolic plans. It then uses AIT* to lazily validate these symbolic plans and find and improve compatible motion plans. This hierarchical, interleaved approach allows TMIT* to share information between all aspects of the TMP search and delay computationally expensive operations until necessary. As a result, TMIT* finds initial solutions quickly and converges toward the optimal integrated task and motion plan in an anytime manner.
- Task and Motion Informed Trees (TMIT*): Almost-surely asymptotically optimal integrated task and motion planning
- IEEE Robotics and Automation Letters (RA-L)
- Presented at IROS 2022