A Simple Approach to Constraint-Aware Imitation Learning with Application to Autonomous Racing

2025-03-10

AI Summary

For safety-critical, high-precision control tasks—such as autonomous racing—conventional imitation learning often fails to enforce hard safety constraints (e.g., physical boundaries and dynamical limits), compromising both safety and performance. To address this, we propose a lightweight, differentiable constraint-embedding mechanism that directly incorporates safety constraints into the behavioral cloning objective function, without requiring auxiliary networks or complex dynamical modeling. Our method supports dual-modality inputs—full-state and raw images—enabling end-to-end training. Evaluated in autonomous racing simulation, it significantly improves constraint satisfaction rates and achieves superior task-performance consistency over baseline methods. Crucially, it demonstrates robustness and generalization across both full-state and vision-based feedback settings. These results validate its effectiveness and broad applicability in safety-critical autonomous systems.

Technology Category

• Machine Learning: Imitation Learning & Inverse Reinforcement Learning
• Constraint Satisfaction and Optimization: Constraint Learning and Acquisition
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems

Abstract

Guaranteeing constraint satisfaction is challenging in imitation learning (IL), particularly in tasks that require operating near a system's handling limits. Traditional IL methods often struggle to enforce constraints, leading to suboptimal performance in high-precision tasks. In this paper, we present a simple approach to incorporating safety into the IL objective. Through simulations, we empirically validate our approach on an autonomous racing task with both full-state and image feedback, demonstrating improved constraint satisfaction and greater consistency in task performance compared to a baseline method.

Problem

• Ensuring constraint satisfaction in imitation learning
• Improving performance in high-precision tasks
• Validating approach on autonomous racing with safety

Innovation

• Incorporates safety into imitation learning objectives
• Validates approach using simulations with feedback
• Improves constraint satisfaction and task consistency