Learning to Correct Mistakes: Backjumping in Long-Horizon Task and Motion Planning

TL;DR

This paper introduces learned backjumping heuristics for long-horizon task and motion planning, significantly improving efficiency over traditional backtracking by directly identifying problematic actions using supervised learning.

Contribution

It presents a novel approach to learn heuristics for backjumping in planning, reducing search time and improving scalability in complex robotic tasks.

Findings

Significant efficiency gains over backtracking methods.

Effective generalization to problems with different object counts.

Successful application across multiple task domains.

Abstract

As robots become increasingly capable of manipulation and long-term autonomy, long-horizon task and motion planning problems are becoming increasingly important. A key challenge in such problems is that early actions in the plan may make future actions infeasible. When reaching a dead-end in the search, most existing planners use backtracking, which exhaustively reevaluates motion-level actions, often resulting in inefficient planning, especially when the search depth is large. In this paper, we propose to learn backjumping heuristics which identify the culprit action directly using supervised learning models to guide the task-level search. Based on evaluations on two different tasks, we find that our method significantly improves planning efficiency compared to backtracking and also generalizes to problems with novel numbers of objects.