Reward-modulated learning using spiking neural networks for vehicle lateral control

TL;DR

This paper introduces a spiking neural network-based vehicle lateral controller that mimics a model-based controller and can adapt online through reward-modulated plasticity, enhancing performance in dynamic driving scenarios.

Contribution

It presents a novel online adaptive vehicle control method using reward-modulated learning in spiking neural networks, combining biological plausibility with practical control performance.

Findings

The SNN controller successfully replicates the baseline controller behavior.

Online adaptation improves control accuracy in complex paths.

Simulation results demonstrate effective learning and adaptation.

Abstract

This paper presents a vehicle lateral controller based on spiking neural networks capable of replicating the behavior of a model-based controller but with the additional ability to perform online adaptation. By making use of neural plasticity and thanks to reward modulation learning, neural connections are modified to adjust the response according to the committed error. Therefore, the error performs a similar role to dopamine in a biological system, modulating the learning process based on spiking time dependency. The connections are initially set to replicate behavior of the model-based controller. Online adaptation allows tuning connection parameters to improve controller performance. A path controller with a preview is used as a baseline controller to evaluate the performance of the proposed approach. Key performance indicators are obtained from simulation with a step response and a handling track with 20 curves of different radii.