AI-Enabled Adaptive Cruise Control for Energy Optimization in Autonomous Highway Driving

Abstract

Energy-aware adaptive cruise control for autonomous highway vehicles needs to consider safety and comfort in a dynamic traffic environment simultaneously with propulsion efficiency. A hierarchical artificial intelligence controller is proposed in this paper that combines a temporal attention predictor, an economic model predictive controller, and a bounded reinforcement-learning residual policy. The predictor estimates a 6s lead-vehicle trajectory and local traffic behaviour; the optimizer generates a constraint-feasible speed profile in line with the grade and powertrain maps; and the residual policy corrects for prediction errors without exceeding the hard headway, acceleration or jerk limits. A high-fidelity co-simulation evaluates 1,440 episodes in flat, rolling, congested, cut-in and communication-degraded highway conditions. Compared with a calibrated production-style adaptive cruise control system, the proposed method has reduced the mean traction energy by 13.7%, regenerative loss by 9.4%, and root-mean-square acceleration by 18.6%, while maintaining a minimum time headway of 1.21 s and a zero-collision record. Compared with economic model predictive control without learned adaptation, it offers an additional 5.2% energy reduction and reduces speed-tracking error by 14.8%. Ablation and sensitivity results show that traffic prediction, grade preview and bounded residual learning are all beneficial to some extent. Based on research, safe reinforcement learning can reduce highway energy consumption by making multiple traffic and road predictions without losing longitudinal stability.