Wei Han

Learning-based active suspension control requires a policy that can adapt to changing road excitations, respect actuator feasibility, and optimize ride comfort in a form consistent with vehicle-vibration evaluation. Existing reinforcement learning controllers for suspension systems are often trained on a narrow set of road profiles, use generic acceleration penalties as rewards, and provide limited evidence on how road-aware state design and comfort-aligned reward design affect cross-road generalization. This study develops RAC-TD3, a road-adaptive comfort-oriented Twin Delayed Deep Deterministic Policy Gradient framework for continuous active suspension control. The proposed formulation uses a feasibility-bounded four-actuator action space, a road-conditioned Markov state that combines vehicle dynamic states with road-excitation descriptors, and a tri-axial weighted comfort reward inspired by ISO 2631 seat-vibration assessment. A 7-degree-of-freedom full-vehicle model with driver-seat dynamics, nonlinear suspension characteristics, ISO stochastic roads, hump excitation, and reconstructed real-road elevation is used for evaluation. Across ISO Class A--F roads, humps, and a KITTI-derived real-road profile, RAC-TD3 achieves the lowest average weighted acceleration RMS among the reinforcement-learning variants, reducing the average RMS by 41.18% relative to baseline TD3. Frequency-domain analysis further shows attenuation near the estimated sprung-body and unsprung wheel-hop frequency bands, indicating that the learned policy suppresses physically meaningful vibration modes rather than only reducing time-domain peaks.

Reinforcement learning (RL) has shown excellent performance in solving decision-making and control problems of autonomous driving (AD), which is increasingly applied in diverse driving scenarios. However, driving is a multiattribute problem, leading to challenges in achieving multiobjective compatibility for current RL methods, especially in both policy updating and policy execution. On the one hand, a single value evaluation network limits the policy updating in complex scenarios with coupled driving objectives. On the other hand, the common single-type action space structure limits driving flexibility or results in large behavior fluctuations during policy execution. To this end, we propose a multiobjective ensemble-critic (MoEC) RL method with a hybrid parametrized action for multiobjective compatible AD. Specifically, an advanced MORL architecture is constructed, in which the ensemble-critic focuses on different objectives through independent reward functions. The architecture integrates a hybrid parameterized action space structure, and the generated driving actions contain both abstract guidance that matches the hybrid road modality and concrete control commands. In addition, an uncertainty-based exploration mechanism that supports hybrid actions is developed to learn multiobjective compatible policies more quickly. The experimental results demonstrate that, in both simulator-based and HighD dataset-based multilane highway scenarios, our method efficiently learns multiobjective compatible AD with respect to efficiency, action consistency, and safety.

Brake-by-wire (BBW) systems significantly enhance vehicle safety through high control accuracy and rapid response, enabled by a smart electromechanical actuator. In this actuator, an electric motor drives a rotational-to-linear reduction mechanism to generate braking force. Most studies employ a pressure feedback framework with pressure sensors to track the desired pressure amid nonlinearities and uncertainties in BBW systems. While the requirements for cost reduction, space constraints, and fail-operational capability for BBW systems makes it essential to investigate the pressure-sensorless control by adopting the motor information of BBW systems. To this end, a novel layer-by-layer progressive(LLP) control methodology is proposed involving coarse compensation, fine compensation and fine control. Coarse compensation is used for modelable nonlinearity: A load-adaptive LuGre model is constructed to handle the static, dynamic and load-dependent friction nonlinearity. Fine compensation is utilized to eliminate the estimable disturbances: An adaptive extended disturbance observer(AEDO) accounts for parameter uncertainty and external disturbances separately. Fine control is employed to attenuate the non-modelable nonlinearities and non-estimable disturbances: A cascade nonlinear controller is designed. Unlike prior studies, this work fully accounts for actuator characteristics and employs differentiated solutions to progressively address system challenges. Hardware-in-the-Loop (HIL) experimental results for the proposed LLP method demonstrate a steady-state error of less than 0.5 bar during normal driver braking. The control deviation and other indicators show that the proposed LLP has improved the sensorless control performance by over 10% superior to the best baseline method.

With the gradual advancement of technology from advanced driver assistance systems (ADAS) to full autonomous driving, integrating multiple functions across different operational design domains (ODDs) has become imperative. Integrated cruising and parking—one of the most critical integrated functions for autonomous electric vehicles—demands a consistent algorithmic framework, particularly for reference path tracking. However, previous studies have rarely explored a unified path tracking control framework capable of adapting to both scenarios. To address the path tracking requirements of integrated cruising and parking, this paper develops a cascade control framework that incorporates both vehicle kinematics and dynamics. The lateral control algorithm is detailed designed based on the kinematics-based sliding mode control with conditional integrator and dynamics-based predictive controller for cruising and parking, respectively. The performance of the proposed framework is validated through real-vehicle experiments under typical integrated cruising and parking situations. Experimental results demonstrate that the maximum lateral error is less than 0.14 m (compared to the industry threshold of 0.3 m) and the maximum heading error is less than 3.5° (versus the industry limit of 8°) in representative test scenarios, confirming its effectiveness and industrial applicability.

Autonomous driving has emerged as a highly topical subject within the realm of intelligent transportation systems. Automated valet parking (AVP) represents one of the initial mass‐production application scenarios. However, motion planning in AVP confronts a series of formidable challenges. These challenges include a constricted movement space, vehicles parked in violation of regulations, and vehicles that intrude suddenly. In response to these issues, this article devises a safety‐critical, kinematically executable overtaking planning system for AVP through a contingency path‐speed iterative algorithm. A path‐speed iterative optimisation framework is adopted, taking into full account both the curvature constraint and the contour constraint. The prediction probability of dynamic obstacles is incorporated into the quadratic optimisation problem, presented in the form of either soft or hard constraints. Furthermore, a contingency path‐speed iterative planner is formulated to address the multi‐modal predictions and the interframe probability transfer that occur during the overtaking process in parking lots. Numerical simulations (conducted on the Carla simulator with a 10 Hz planning cycle) across four complex AVP scenarios demonstrate that the proposed algorithm outperforms the baseline Baidu Apollo EM Planner. On‐road experiments (deployed on a mass‐produced MCU) further validate that the algorithm maintains real‐time performance (average computation time < 10 ms) and reduces speed oscillation by over 50% compared to the baseline, while ensuring kinematically executable trajectories (max steering wheel angle limited to 389°). These results confirm the proposed algorithm significantly enhances overtaking safety, executability, and efficiency for AVP.