Faster Model Predictive Control via Self-Supervised Initialization Learning

2024-08-06

arXiv.org

AI Summary

To address the slow convergence and poor generalization of non-convex, long-horizon model predictive control (MPC) in real-time robotic control, this paper proposes a synergistic offline-online optimization framework comprising offline self-supervised initial-value learning and online gradient-based fine-tuning. We introduce the first self-supervised learning paradigm specifically designed for MPC initialization—requiring neither ground-truth labels nor expert demonstrations—enabling zero-shot transfer to unseen tracks. This approach overcomes the dual bottlenecks of conventional single-stage training in both computational speed and generalization capability. Evaluated on a Formula 1 trajectory-tracking task, our method reduces average MPC optimization time by 19.4% and improves tracking accuracy by 6.3%, significantly enhancing the real-time performance and cross-scenario adaptability of non-convex MPC.

Technology Category

• Search and Optimization: Learning to Search
• Intelligent Robots: Learning & Optimization for ROB
• Machine Learning: Online Learning & Bandits

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Optimization for robot control tasks, spanning various methodologies, includes Model Predictive Control (MPC). However, the complexity of the system, such as non-convex and non-differentiable cost functions and prolonged planning horizons often drastically increases the computation time, limiting MPC's real-world applicability. Prior works in speeding up the optimization have limitations on optimizing MPC running time directly and generalizing to hold out domains. To overcome this challenge, we develop a novel framework aiming at expediting optimization processes directly. In our framework, we combine offline self-supervised learning and online fine-tuning to improve the control performance and reduce optimization time. We demonstrate the success of our method on a novel and challenging Formula 1 track driving task. Comparing to single-phase training, our approach achieves a 19.4% reduction in optimization time and a 6.3% improvement in tracking accuracy on zero-shot tracks.

Problem

• Reduces computation time in Model Predictive Control
• Addresses non-convex and non-differentiable cost functions
• Improves control performance and generalization to new domains

Innovation

• Self-supervised learning for initialization
• Online fine-tuning for optimization
• Reduced computation time in MPC