PALM: Progress-Aware Policy Learning via Affordance Reasoning for Long-Horizon Robotic Manipulation
🤖 AI Summary
Existing vision–language–action models struggle to identify task-relevant interaction cues and track subtask progress in long-horizon, multi-step robotic tasks, often leading to errors such as repetition, omission, or premature termination. To address this, this work proposes an interaction-centric policy learning framework that unifies perception and execution by integrating an affordance representation encompassing object relevance, contact geometry, spatial layout, and motion dynamics, along with a continuous subtask progress prediction mechanism. The proposed approach achieves a 91.8% success rate on LIBERO-LONG, increases the average episode length by 12.5% on the CALVIN ABC→D benchmark, and demonstrates a twofold performance improvement across three real-world long-horizon generalization tasks.
Application Category
📝 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.
Problem
Research questions and friction points this paper is trying to address.
long-horizon manipulation
multi-step tasks
progress tracking
affordance reasoning
execution errors
Innovation
Methods, ideas, or system contributions that make the work stand out.
affordance reasoning
progress-aware policy learning
long-horizon manipulation
vision-language-action models
subtask progress tracking
🔎 Similar Papers
No similar papers found.
