Unified Task and Motion Planning using Object-centric Abstractions of Motion Constraints

TL;DR

This paper tackles the difficulty of integrating task and motion planning under physical constraints by introducing U-TAMP, a unified heuristic search framework built on object-centric abstractions of motion constraints. By representing grasping, placement, and kinematic relations through object parts and bounding-box surfaces, the approach enables task planners to reason about motion feasibility directly, avoiding costly sub-symbolic geometry during planning. The method defines ground predicates for two task families—object-support and object-container—along with corresponding PDDL actions that yield geometrically consistent plans that can be executed without further geometric reasoning. Experimental results on green-block rearrangement and a cooking-for-two scenario show substantial speedups over state-of-the-art TAMP methods, highlighting the practical impact of object-centric abstractions for scalable, robust long-horizon manipulation. Overall, the work demonstrates that unifying task and motion planning at the symbolic level with carefully designed object-centric constraints can significantly reduce planning time while ensuring feasible, executable plans.

Abstract

In task and motion planning (TAMP), the ambiguity and underdetermination of abstract descriptions used by task planning methods make it difficult to characterize physical constraints needed to successfully execute a task. The usual approach is to overlook such constraints at task planning level and to implement expensive sub-symbolic geometric reasoning techniques that perform multiple calls on unfeasible actions, plan corrections, and re-planning until a feasible solution is found. We propose an alternative TAMP approach that unifies task and motion planning into a single heuristic search. Our approach is based on an object-centric abstraction of motion constraints that permits leveraging the computational efficiency of off-the-shelf AI heuristic search to yield physically feasible plans. These plans can be directly transformed into object and motion parameters for task execution without the need of intensive sub-symbolic geometric reasoning.

Figures (15)

• Figure 1: 3D rotation in terms of roll, pitch, and yaw. Source lavalle2006planning.
• Figure 2: TAMP templates for pick-and-place actions adapted from garrett2021integrated.
• Figure 3: Schema of hand-object configurations for four different grasp types.
• Figure 4: Example of four placement types, as described in Table \ref{['table:example_place_types']}.
• Figure 5: Example hand-object and object-support configurations in the kinematics constraint.