Incremental Few-Shot Adaptation for Non-Prehensile Object Manipulation using Parallelizable Physics Simulators

2024-09-20

arXiv.org

AI Summary

This work addresses the few-shot adaptation problem for non-grasping object manipulation in open-world settings, enabling real-time model predictive control (MPC). We propose an incremental dynamics model adaptation method that integrates parallelized rigid-body physics simulation as a differentiable dynamic world model, coupled with sampling-based optimization—specifically the cross-entropy method (CEM)—to enable few-shot, incremental calibration of dynamics parameters. Crucially, our approach requires only 3–5 real-robot interactions to adapt to previously unseen objects—including novel geometries and friction properties. Evaluated on both simulation and physical pushing tasks, it improves task success rates by over 40% compared to baseline MPC methods. The method significantly enhances the generalization capability and online adaptation speed of MPC in open, unstructured environments, marking the first use of differentiable, parallel physics simulation for such rapid dynamics identification in robotic manipulation.

Technology Category

• Intelligent Robots: Manipulation
• Search and Optimization: Sampling/Simulation-based Search
• Cognitive Modeling & Cognitive Systems: Adaptive Behavior

Application Category

• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• 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

Abstract

Few-shot adaptation is an important capability for intelligent robots that perform tasks in open-world settings such as everyday environments or flexible production. In this paper, we propose a novel approach for non-prehensile manipulation which incrementally adapts a physics-based dynamics model for model-predictive control (MPC). The model prediction is aligned with a few examples of robot-object interactions collected with the MPC. This is achieved by using a parallelizable rigid-body physics simulation as dynamic world model and sampling-based optimization of the model parameters. In turn, the optimized dynamics model can be used for MPC using efficient sampling-based optimization. We evaluate our few-shot adaptation approach in object pushing experiments in simulation and with a real robot.

Problem

• Adapting physics models for non-prehensile manipulation tasks
• Aligning model predictions with few interaction examples
• Optimizing dynamics for model-predictive control in robotics

Innovation

• Incremental few-shot adaptation for manipulation
• Parallelizable physics simulation as dynamic model
• Sampling-based optimization for model parameters