Task and Motion Informed Trees (TMIT*): Almost-Surely Asymptotically Optimal Integrated Task and Motion Planning

High-level autonomy requires discrete and continuous reasoning to decide both what actions take how execute them. Integrated Task Motion Planning (TMP) algorithms solve these hybrid problems jointly consider constraints between the symbolic (i.e., task plan) their geometric realization motion plans). This joint approach solves more difficult than approaches that address subproblems independently. TMP combine extend results from planning. has mainly focused on computational performance completeness less solution optimality. Optimal is because independent optima of may not be optimal integrated solution, which can only found by optimizing plans. paper presents Informed Trees (TMIT*), an algorithm combines makespan-optimal planning almost-surely asymptotically TMIT* interleaves asymmetric forward reverse searches delay computationally expensive operations until necessary perform efficient informed search directly in problem's state space. allows it quickly then converge towards with additional time, as demonstrated evaluated robotic-manipulation benchmark problems.