Macro-Operator Generation and Predicate Selection for TAMP Operator Learning

This paper introduces a system that automatically generates macro-operators by discovering causally linked action pairs from demonstration data, while simultaneously pruning irrelevant predicates to reduce the symbolic state. The approach addresses bottlenecks in Task and Motion Planning (TAMP) where existing methods learn actions in isolation or evaluate unnecessary predicates. By compressing recurring multi-step manipulation sequences into single planning steps and shrinking the search space, the method achieves up to a ~4.6x speedup over the baseline 'Learning Operators for TAMP' across four domains. Crucially, it enables solving long sequential tasks that the baseline fails to solve, demonstrating that macro-operator discovery can determine practical solvability in long-horizon scenarios. The code is available at https://github.com/canemirbora4/macro-tamp.

Generated from attributed reports · Updated 1 hours ago

There is not enough continuous observation data to show a trend.