PDDLStream: Integrating Symbolic Planners and Blackbox Samplers via Optimistic Adaptive Planning

Abstract

Many planning applications involve complex relationships defined on high-dimensional, continuous variables. For example, robotic manipulation requires planning with kinematic, collision, visibility, and motion constraints involving robot configurations, object poses, and robot trajectories. These constraints typically require specialized procedures to sample satisfying values. We extend PDDL to support a generic, declarative specification for these procedures that treats their implementation as black boxes. We provide domain-independent algorithms that reduce PDDLStream problems to a sequence of finite PDDL problems. We also introduce an algorithm that dynamically balances exploring new candidate plans and exploiting existing ones. This enables the algorithm to greedily search the space of parameter bindings to more quickly solve tightly-constrained problems as well as locally optimize to produce low-cost solutions. We evaluate our algorithms on three simulated robotic planning domains as well as several real-world robotic tasks.

Conflict-Directed Diverse Planning for Logic-Geometric Programming

“…Similarly, the method in (Dantam et al 2016) incrementally incorporates information about the motion feasibility into the symbolic description by using a constraint-based task planning formulation and a Satisfiability Modulo theory (SMT) solver. Another related approach is PDDLStream (Garrett, Lozano-Pérez, and Kaelbling 2020), which also combines symbolic planners with motion planning. Specifically, PDDLStream uses constrained samplers (in configuration space) to discretize the motion planning problem, and PDDL planning to solve the combined problem.…”

“…Therefore, we do not compare to other methods in task and motion planning, e.g. (Garrett, Lozano-Pérez, and Kaelbling 2020), that use different underlying problem formulations and methods.…”

“…Combined task and motion planning (Garrett et al 2021) combines both symbolic planning about the task with trajectory planning for the motions involved with the plan. Many approaches to combined task and motion planning have been developed (Srivastava et al 2014;Bidot et al 2017;Ferrer-Mestres, Francès, and Geffner 2017;Garrett, Lozano-Pérez, and Kaelbling 2020;Lagriffoul et al 2014) over the years.…”

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

“…Schmitt et al [16] offer almost-surely asymptotically optimal manipulation planning based on a precomputed roadmap and domain-specific samplers for mode transitions. Garrett et al [17] produce TMP solutions that are optimal in symbolic action cost, but assume domain-specific samplers and do not jointly optimize their motions and action parameter choices. Recent results [5] have shown that almost-sure asymptotic optimality results from pure motion planning are preserved for multimodal TMP under realistic assumptions on the measure of mode transition sets.…”

“…TMP solvers adapt advances in standalone symbolic planning to the TMP context [1,17,19,20]. Symbolic planning for TMP is uniquely challenging since valid symbolic plans may not correspond to valid motion plans.…”