Multi-Objective Reinforcement Learning With Physics-Aware Vehicle Dynamics for Safe and Efficient Adaptive Cruise Control

Title

Multi-Objective Reinforcement Learning With Physics-Aware Vehicle Dynamics for Safe and Efficient Adaptive Cruise Control

Creator

Haneesh, K.M.; Jisha, P.

Description

Adaptive Cruise Control (ACC) enhances safety and comfort in autonomous vehicles by maintaining appropriate inter-vehicular distance and speed regulation. Traditional ACC systems based on PID or Model Predictive Control (MPC) often struggle to handle complex, unforeseen traffic scenarios such as sudden braking, pedestrian crossings, or lane changes. Reinforcement Learning (RL) offers an adaptive alternative by enabling policy learning through environment interaction. However, existing RL-based ACC methods frequently suffer from poor smoothness and energy inefficiency under emergency conditions. This work proposes an enhanced RL-based ACC framework that integrates a physics-informed, multi-objective reward function to jointly optimize safety, ride comfort, and energy efficiency. The reward components are normalized and dynamically weighted based on the current driving context, allowing the agent to adaptively prioritize objectives. Vehicular dynamics are explicitly incorporated into the learning process to improve real-world applicability. The system is trained using the DDPG algorithm, which supports continuous control and stable policy convergence. Extensive MATLAB-based simulations were conducted across diverse urban driving scenarios including stopgo traffic, traffic signals, lane changes, and pedestrian interactions. Comparative analysis against PID and MPC-based ACC controllers demonstrates that the proposed framework achieves superior performance in maintaining safe inter-vehicular distance, reducing jerk, and improving energy efficiency. This study validates the feasibility of deploying a computationally efficient, model-free RL-based ACC for robust and safe autonomous driving in dynamic traffic environments. The Author(s), under exclusive licence to ITS Japan 2026.

Source

International Journal of Intelligent Transportation Systems Research;

Date

01-01-2026

Publisher

Springer

Subject

Adaptive cruise control; Deep Deterministic Policy Gradient; Multi-objective reward function; Physics-aware vehicle dynamics; Reinforcement learning

Coverage

Haneesh K.M., Department of Electrical and Electronics Engineering, CHRIST University, Bangalore, India; Jisha P., Department of Electronics and Communication Engineering, BMS College of Engineering, Bangalore, India

Rights

Restricted Access; Hardcopy may be available in the library

Relation

ISSN: 13488503;

Format

online

Language

English

Type

Article

Identifier

https://doi.org/10.1007/s13177-026-00669-y

https://www.scopus.com/pages/publications/105039476332?origin=resultslist