Hybrid DDPG-LQR control for intelligent active suspension with real-time adaptive fusion

This is the authors' abstract. We don't add key points for this paper.

Traditional active suspension control faces the challenge of jointly balancing ride comfort, driving safety, and suspension travel utilization. This paper proposes a hybrid control framework that integrates the Deep Deterministic Policy Gradient (DDPG) controller with a Linear Quadratic Regulator (LQR). Using a quarter-car dynamic model, an LQR benchmark is established alongside a DDPG reinforcement learning framework. A novel multi-objective adaptive reward function is designed to co-optimize vertical body acceleration, suspension deflection, and dynamic tire load. Crucially, to bridge these two distinct controllers, we introduce a real-time, performance-based adaptive fusion mechanism that dynamically weights LQR and DDPG to combine stability and adaptability. Simulations demonstrate that the hybrid approach achieves a higher overall performance score than standalone LQR or DDPG, improving over passive suspension by 26.9%, reducing suspension travel by 33.3%, and decreasing energy consumption by 20.7% relative to the LQR baseline, while exhibiting high robustness against extreme perturbations and effective transferability to a full-vehicle environment. Experiments verify the algorithm’s stable force tracking.