ActivePusher: Active Learning and Planning with Residual Physics for Nonprehensile Manipulation

2025-06-05

AI Summary

For long-horizon motion planning in non-grasping manipulation tasks (e.g., pushing, rolling), efficient dynamics modeling remains challenging due to the high sample cost of learning-based approaches. This paper proposes a joint framework integrating residual physics-informed modeling with kernel-based uncertainty quantification and active learning. Specifically, it performs information-directed data acquisition in the skill parameter space, explicitly incorporating Gaussian-kernel-estimated model uncertainty into the sampling procedure of Model Predictive Control (MPC). This enhances data efficiency by prioritizing informative queries. Evaluated in simulation and on real robotic platforms, the method reduces required interaction samples by over 40% compared to random sampling baselines, while improving planning success rate by more than 25%. These results demonstrate substantial alleviation of the performance bottleneck imposed by inefficient data collection in learned dynamics models.

Technology Category

• Intelligent Robots: Motion and Path Planning
• Planning, Routing, and Scheduling: Learning for Planning and Scheduling
• Machine Learning: Calibration & Uncertainty Quantification

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Planning with learned dynamics models offers a promising approach toward real-world, long-horizon manipulation, particularly in nonprehensile settings such as pushing or rolling, where accurate analytical models are difficult to obtain. Although learning-based methods hold promise, collecting training data can be costly and inefficient, as it often relies on randomly sampled interactions that are not necessarily the most informative. To address this challenge, we propose ActivePusher, a novel framework that combines residual-physics modeling with kernel-based uncertainty-driven active learning to focus data acquisition on the most informative skill parameters. Additionally, ActivePusher seamlessly integrates with model-based kinodynamic planners, leveraging uncertainty estimates to bias control sampling toward more reliable actions. We evaluate our approach in both simulation and real-world environments and demonstrate that it improves data efficiency and planning success rates compared to baseline methods.

Problem

• Improving data efficiency in nonprehensile manipulation planning
• Reducing training costs via uncertainty-driven active learning
• Enhancing planning success with residual-physics modeling integration

Innovation

• Combines residual-physics with active learning
• Uses kernel-based uncertainty-driven data acquisition
• Integrates with model-based kinodynamic planners