AutomaChef: A Physics-informed Demonstration-guided Learning Framework for Granular Material Manipulation

2024-06-13

arXiv.org

AI Summary

Particle manipulation faces challenges in accurate physical modeling, low-fidelity simulation, and inefficient policy learning. Method: We develop a high-fidelity, differentiable physics simulator built on Taichi, enabling end-to-end coupling of precise particle-system modeling and policy optimization. We propose a cross-material demonstration-guided approach: imperfect demonstrations—generated via gradient-based optimization on non-granular materials—are leveraged to accelerate granular policy training. Furthermore, we integrate physics-informed neural networks with imitation-reinforcement learning to achieve seamless sim-to-real transfer. Contribution/Results: On a granular transport task, we successfully deploy three cascaded policies. Both simulation and real-world performance significantly surpass mainstream deep reinforcement learning baselines. Our framework overcomes the limitations of conventional methods that rely on low-fidelity surrogate models or neglect fundamental physical principles.

Technology Category

• Machine Learning: Imitation Learning & Inverse Reinforcement Learning
• Search and Optimization: Sampling/Simulation-based Search
• Computer Vision: Low Level & Physics-based Vision

Application Category

• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments

Abstract

Due to the complex physical properties of granular materials, research on robot learning for manipulating such materials predominantly either disregards the consideration of their physical characteristics or uses surrogate models to approximate their physical properties. Learning to manipulate granular materials based on physical information obtained through precise modelling remains an unsolved problem. In this paper, we propose to address this challenge by constructing a differentiable physics simulator for granular materials based on the Taichi programming language and developing a learning framework accelerated by imperfect demonstrations that are generated via gradient-based optimisation on non-granular materials through our simulator. Experimental results show that our method trains three policies that, when chained, are capable of executing the task of transporting granular materials in both simulated and real-world scenarios, which existing popular deep reinforcement learning models fail to accomplish.

Problem

• Developing robot learning for granular material manipulation using physical properties
• Creating differentiable physics simulator to model complex granular material dynamics
• Training robust policies for granular material transport in simulated and real environments

Innovation

• Differentiable physics simulator using Taichi language
• Demonstration-guided learning with gradient optimization
• Flexible framework for granular material manipulation