Deep Deterministic Path Following

TL;DR

This paper applies the DDPG algorithm to autonomous vehicle path following, demonstrating that the agent can learn policies that minimize cross-track and velocity errors in a simulated environment.

Contribution

It introduces a deep reinforcement learning approach using DDPG for vehicle path following in a custom simulation environment.

Findings

The DDPG agent achieves small cross-track errors.

The agent adapts acceleration to reduce velocity error.

The approach demonstrates effective policy learning for vehicle control.

Abstract

This paper deploys the Deep Deterministic Policy Gradient (DDPG) algorithm for longitudinal and lateral control of a simulated car to solve a path following task. The DDPG agent was implemented using PyTorch and trained and evaluated on a custom kinematic bicycle environment created in Python. The performance was evaluated by measuring cross-track error and velocity error, relative to a reference path. Results show how the agent can learn a policy allowing for small cross-track error, as well as adapting the acceleration to minimize the velocity error.