Physics-guided online adaptive vehicle dynamics modeling via decoupled residual learning

Physics-Informed Imitation Learning for Resilient Autonomous Driving

Imitation learning has emerged as a promising paradigm for autonomous vehicle control. However, existing policies often suffer from severe covariate shift and lack explicit stability guarantees in closed-loop trajectory tracking. This paper proposes an adaptive Lyapunov-constrained imitation learning framework for vehicle trajectory tracking. To address the covariate shift problem, we explicitly embed a discrete-time kinematic bicycle model into the policy training process. This differentiable physical prior enables propagation of the nominal tracking error and allows a global Lyapunov decay constraint to be directly employed on the policy optimization via an Actor-Critic architecture. To prevent the overly aggressive penalization near the reference trajectory, we introduce a state-dependent adaptive decay rate that relaxes the constraint for small errors while strengthening it under large deviations. As a result, the proposed method achieves both high tracking precision and strong recovery from model uncertainties and disturbances. Extensive experiments in interactive traffic environments demonstrate that the proposed method significantly outperforms state-of-the-art baselines in collision avoidance and disturbance rejection.

Vehicle Multimodal Trajectory Prediction Integrating Kinematics and Dynamic Interaction Features

A Stochastic Gating Decoder for multimodal latent variable sampling, adaptively fusing kinematics and data-driven paths to capture driver intention uncertainty while maintaining kinematic consistency is introduced.

Meta-Learning Based USV Dynamics Residual Modeling for MPC Trajectory Tracking under Different Sea States

Unmanned Surface Vehicles (USVs) encounter varying disturbances across different sea states, making it difficult to achieve stable trajectory tracking based solely on nominal physical models. This limitation has motivated the development of disturbance observer-based compensation schemes. However, existing disturbance observers often suffer from slow updating rates, resulting in insufficient adaptability. Therefore, this paper presents a meta-learning based dynamic residual modeling method. To facilitate rapid adaptation to disturbances, Model-Agnostic Meta-Learning is employed for pre-training. Through dual-loop gradient updates, the meta-learned initial parameters for the residual dynamics model are obtained. Subsequently, the residual model is integrated with the nominal model and embedded into the model predictive control framework, utilizing few-shot observation data to execute online fine-tuning of the residual network, thereby continuously correcting the dynamic residuals induced by time-varying sea states. Simulation experiments conducted in Gazebo VRX demonstrate that the proposed method maintains stable trajectory tracking in time-varying environments, significantly enhancing the overall disturbance rejection capability and robustness of the USV in complex sea states.

Stable Multi-Step Rollouts via Uncertainty-Guided Hybrid Dynamics

Multi-step rollouts are essential for model-based reinforcement learning (RL) and predictive control, yet learned dynamics models often become unstable when recursively applied, leading to divergence and unreliable policy updates. This paper proposes a model-agnostic hybrid dynamics framework that blends a provably contracting nominal model with a flexible excursion model through an uncertainty-guided switching law. The switching signal is derived from calibrated epistemic uncertainty and activates only when the system leaves the nominal region, ensuring that each model operates within its reliability regime. Under clearly stated smoothness and boundedness assumptions, we show that the resulting hybrid predictor yields globally bounded recursive multi-step rollouts: trajectories remain Lyapunov-stable in the nominal region and exhibit at most affine growth during excursions. To illustrate the theory in practice, we instantiate the hybrid dynamics framework within a model-based RL scheme that uses real one-step transitions for value learning and hybrid rollouts for policy improvement. Experiments on a nonlinear Duffing oscillator demonstrate stable long-horizon prediction and improved cost-effort trade-offs relative to a stabilizing baseline.

Data-driven stochastic model predictive control for spacecraft attitude tracking

Incorporating deep operator network into feedforward–feedback controller synthesis for path tracking of autonomous vehicles

In this paper, a deep operator network (DeepONet)-based feedforward–feedback control framework for autonomous vehicles’ path tracking controller is introduced. In the proposed framework, DeepONet is utilized for data-driven vehicle dynamics modeling within the feedforward control component. To improve the accuracy of lateral dynamics, the physical model is integrated into the DeepONet architecture as prior knowledge. By incorporating experimentally validated physical models, the proposed approach captures the causal relationship between vehicle parameters (e.g. mass, tire-road friction) and vehicle dynamics, thereby enhancing the interpretability of the model. In the CarSim/Simulink simulation environment, the designed DeepONet-based feedforward–feedback controller is evaluated in two scenarios: double-lane change and oval track. The validation results demonstrate that the proposed approach achieves smaller lateral errors than other methods in both linear and nonlinear regions. Notably, when facing unknown roads and varying road friction conditions on the oval track, the DeepONet-based approach shows improvement in tracking accuracy compared to purely data-driven methods without physical model integration.