Emerging Extrinsic Dexterity in Cluttered Scenes via Dynamics-aware Policy Learning

2026-03-10

AI Summary

This work addresses the challenge of non-prehensile manipulation in cluttered environments, where selective exploitation of dynamic effects amidst multi-object coupled contacts is required—a task for which existing methods lack explicit modeling of complex interactions. To this end, we propose the Dynamics-Aware Policy Learning (DAPL) framework, which learns an explicit world model to represent contact-induced object dynamics and leverages this representation to guide conditional reinforcement learning. This approach enables the emergent acquisition of extrinsic dexterity without handcrafted heuristics or finely tuned reward functions. DAPL achieves the first end-to-end learned dexterous manipulation in clutter by explicitly modeling contact dynamics, outperforming prior methods by over 25% in success rate on unseen simulated scenarios. In real-world experiments across ten cluttered scenes, it attains an average success rate of approximately 50% and demonstrates strong sim-to-real transferability, validated through deployment in a real supermarket setting.

Technology Category

• Intelligent Robots: Manipulation
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Machine Learning: Imitation Learning & Inverse Reinforcement Learning

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

Abstract

Extrinsic dexterity leverages environmental contact to overcome the limitations of prehensile manipulation. However, achieving such dexterity in cluttered scenes remains challenging and underexplored, as it requires selectively exploiting contact among multiple interacting objects with inherently coupled dynamics. Existing approaches lack explicit modeling of such complex dynamics and therefore fall short in non-prehensile manipulation in cluttered environments, which in turn limits their practical applicability in real-world environments. In this paper, we introduce a Dynamics-Aware Policy Learning (DAPL) framework that can facilitate policy learning with a learned representation of contact-induced object dynamics in cluttered environments. This representation is learned through explicit world modeling and used to condition reinforcement learning, enabling extrinsic dexterity to emerge without hand-crafted contact heuristics or complex reward shaping. We evaluate our approach in both simulation and the real world. Our method outperforms prehensile manipulation, human teleoperation, and prior representation-based policies by over 25% in success rate on unseen simulated cluttered scenes with varying densities. The real-world success rate reaches around 50% across 10 cluttered scenes, while a practical grocery deployment further demonstrates robust sim-to-real transfer and applicability.

Problem

• extrinsic dexterity
• cluttered scenes
• non-prehensile manipulation
• contact dynamics
• object interaction

Innovation

• Extrinsic dexterity
• Dynamics-aware policy learning
• Contact-induced dynamics
• Cluttered manipulation
• Sim-to-real transfer