Optimizing Low-Speed Autonomous Driving: A Reinforcement Learning Approach to Route Stability and Maximum Speed

TL;DR

The paper tackles maintaining maximum speed stability in low-speed autonomous driving while following predefined routes, proposing a reinforcement-learning framework with a composite reward that balances speed, progress, and trajectory smoothness. It introduces a max-speed constraint at $1\,\mathrm{m/s}$, progress-based rewards with an epsilon-regularized denominator to address numerical instability, and a curvature-aware steering penalty integrated into a track-aware composite reward. Key contributions include the max-speed control, stability-enhanced progress rewards, curvature-aware steering penalties, and a dynamic, segment-aware implementation that improves speed stability and route fidelity. Experimental validation on the AWS DeepRacer platform demonstrates robust performance improvements over traditional methods and results in a competition victory, highlighting practical potential for smoother, more reliable low-speed autonomous driving in constrained environments.

Abstract

Autonomous driving has garnered significant attention in recent years, especially in optimizing vehicle performance under varying conditions. This paper addresses the challenge of maintaining maximum speed stability in low-speed autonomous driving while following a predefined route. Leveraging reinforcement learning (RL), we propose a novel approach to optimize driving policies that enable the vehicle to achieve near-maximum speed without compromising on safety or route accuracy, even in low-speed scenarios.

Figures (6)

• Figure 1: Speed
• Figure 2: Speed
• Figure 3: Progress-Based
• Figure 4: Steering Penalty Rewards
• Figure 5: Comparison Of Steering Penalty Rewards