ActivePusher: Active Learning and Planning with Residual Physics for Nonprehensile Manipulation
Zhuoyun Zhong, Seyedali Golestaneh, Constantinos Chamzas
AI summary
Problem
Accurately modeling dynamics for nonprehensile manipulation is hindered by sample inefficiency and high uncertainty in unexplored skill regions, which undermines reliable long-horizon planning.
Approach
ACTIVEPUSHER combines residual physics neural networks with Neural Tangent Kernel-based uncertainty estimation to actively select informative data for training and bias planning toward high-confidence actions.
Key results
- Active learning framework that maximizes expected information gain for skill model training
- Uncertainty-aware kinodynamic planner that biases action sampling toward reliable regions
- Empirical validation across simulation and real-world pushing tasks with multiple objects
- Significantly improved data efficiency and higher planning success rates compared to baselines
Why it matters
Provides a practical, data-efficient pathway for deploying robust nonprehensile manipulation skills on physical robots without relying on expensive simulators or large datasets.
Abstract
Planning with learned dynamics models offers a promising approach toward versatile real-world manipulation, particularly in nonprehensile settings such as pushing or rolling, where accurate analytical models are difficult to obtain. However, collecting training data for learning-based methods can be costly and inefficient, as it often relies on randomly sampled interactions that are not necessarily the most informative. Furthermore, learned models tend to exhibit high uncertainty in underexplored regions of the skill space, undermining the reliability of long-horizon planning. To address these challenges, we propose ACTIVEPUSHER, a novel framework that combines residual-physics modeling with uncertainty-based active learn- ing, 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 consistently improves data efficiency and achieves higher planning success rates in comparison to baseline methods. The source code is available at https://github.com/elpis-lab/ActivePusher.