Modelling, Positioning, and Deep Reinforcement Learning Path Tracking Control of Scaled Robotic Vehicles: Design and Experimental Validation

This academic paper focuses on the control of mobile robotic systems, specifically scaled robotic cars. The authors propose a two-stage least-square approach for parameter identification of the longitudinal and lateral dynamics of the robot car. They also suggest using a no-reset federated Kalman filter to improve positioning, and an expert demonstrator to speed up the training phase and mitigate the simulation-to-reality gap. The proposed deep reinforcement learning-based path tracking strategy outperforms model-based control strategies and the demonstrator, as confirmed by experimental results.

Publication date: 2024-01-11
Project Page: N/A
Paper: https://arxiv.org/pdf/2401.05194