Distilling Contact Planning for Fast Trajectory Optimization in Robot Air Hockey

2024-07-04

AI Summary

Real-time trajectory optimization for robotic air hockey faces challenges including discontinuous contact dynamics, difficulty in long-horizon reasoning, and the dual requirement of maximizing puck velocity while ensuring striking accuracy. Method: We propose a hierarchical control framework: (i) an upper layer that distills a stochastic optimal control policy via reinforcement learning for robust contact planning; and (ii) a lower layer employing nonlinear model predictive control (NMPC), integrating a multi-rigid-body dynamics model with real-time visual feedback to enforce physical and operational constraints in low-level motion planning. Contribution/Results: This is the first work to jointly exploit rebound-based striking mechanics and manipulator structural properties on a real air-hockey platform, operating near kinematic and dynamic limits to balance speed and precision. Experiments demonstrate superior performance over pure learning- or pure control-based baselines in both simulation and hardware, achieving a 23% improvement in shot success rate and an average system response latency under 45 ms.

Technology Category

• Intelligent Robots: Learning & Optimization for ROB
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Search and Optimization: Learning to Search

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Robot control through contact is challenging as it requires reasoning over long horizons and discontinuous system dynamics. Highly dynamic tasks such as Air Hockey additionally require agile behavior, making the corresponding optimal control problems intractable for planning in realtime. Learning-based approaches address this issue by shifting computationally expensive reasoning through contacts to an offline learning phase. However, learning low-level motor policies subject to kinematic and dynamic constraints can be challenging if operating in proximity to such constraints is desired. This paper explores the combination of distilling a stochastic optimal control policy for high-level contact planning and online model-predictive control for low-level constrained motion planning. Our system learns to balance shooting accuracy and resulting puck speed by leveraging bank shots and the robot's kinematic structure. We show that the proposed framework outperforms purely control-based and purely learning-based techniques in both simulated and real-world games of Robot Air Hockey.

Problem

• Optimizing robot control for dynamic contact tasks like Air Hockey
• Balancing realtime planning with kinematic and dynamic constraints
• Combining learning and control for improved accuracy and speed

Innovation

• Combines stochastic optimal control with model-predictive control
• Learns high-level contact planning offline
• Balances shooting accuracy and puck speed