PALM: Progress-Aware Policy Learning via Affordance Reasoning for Long-Horizon Robotic Manipulation

TL;DR

PALM introduces a vision-language-action framework that enhances long-horizon robotic manipulation by reasoning about affordances and tracking subtask progress, leading to significant improvements in success rates and task completion metrics.

Contribution

It presents a novel affordance reasoning and progress prediction approach that stabilizes and improves long-horizon policy execution in robotic manipulation.

Findings

Achieved 91.8% success on LIBERO-LONG benchmark.

Improved average task length by 12.5% on CALVIN ABC->D.

Doubled real-world performance over baseline methods.

Abstract

Recent advancements in vision-language-action (VLA) models have shown promise in robotic manipulation, yet they continue to struggle with long-horizon, multi-step tasks. Existing methods lack internal reasoning mechanisms that can identify task-relevant interaction cues or track progress within a subtask, leading to critical execution errors such as repeated actions, missed steps, and premature termination. To address these challenges, we introduce PALM, a VLA framework that structures policy learning around interaction-centric affordance reasoning and subtask progress cues. PALM distills complementary affordance representations that capture object relevance, contact geometry, spatial placements, and motion dynamics, and serve as task-relevant anchors for visuomotor control. To further stabilize long-horizon execution, PALM predicts continuous within-subtask progress, enabling seamless subtask transitions. Across extensive simulation and real-world experiments, PALM consistently outperforms baselines, achieving a 91.8% success rate on LIBERO-LONG, a 12.5% improvement in average length on CALVIN ABC->D, and a 2x improvement over real-world baselines across three long-horizon generalization settings.