Hybrid Contact Dynamics and Residual-RL Framework for Multi-Point Object Pushing
Chen Chen, Xu Dai, Jozsef Kovecses
AI summary
Problem
Robotic pushing is hindered by complex frictional interactions and nonlinear dynamics, which are often oversimplified by quasi-static models or ignored by data-heavy learning methods.
Approach
The framework integrates a physics-based dynamic model using unilateral constraints and box friction with an RL residual module that fine-tunes end-effector velocity commands based on a force-level reward function.
Key results
- Formulated a comprehensive dynamic contact model for redundant robotic arms
- Extended the model to accommodate multiple simultaneous point contacts
- Developed a residual RL module to correct friction uncertainties and contact disturbances
- Achieved more accurate trajectory following than traditional PD control in real-world Kinova Gen2 experiments
Why it matters
This method allows robots without grippers or those handling heavy objects to achieve stable, high-precision manipulation by bridging analytical modeling and adaptive learning.
Abstract
Robotic contact manipulation involves applying con- trolled forces at contact points to guide an object along a desired trajectory while respecting the underlying physical interactions. This letter presents a novel framework that integrates dynamic modeling and Reinforcement Learning (RL) to achieve robust ob- ject pushing with a redundant robotic arm. First, a comprehensive dynamic contact model is formulated, incorporating unilateral constraints and a box friction model to capture the nonlineari- ties present in real-world contact dynamics. Second, the model is extended to handle multiple simultaneous point contacts, enabling effective trajectory planning and tracking for a redundant robotic arm in multi-contact pushing tasks. Third, an RL strategy is intro- duced as a residual module that augments a model-based controller to improve pushing performance. Simulation and real-world ex- periments with a Kinova Gen2 arm demonstrate that the proposed method achieves accurate trajectory following and stable contact interactions, significantly outperforming traditional PD control strategies in dynamic pushing scenarios.