The skill to drift a car-i.e., operate in a state of controlled oversteer like professional drivers-could give future autonomous cars maximum flexibility when they need to retain control in adverse conditions or avoid collisions. We investigate real-time drifting strategies that put the car where needed while bypassing expensive trajectory optimization. To this end, we design a reinforcement learning agent that builds on the concept of tire energy absorption to autonomously drift through changing and complex waypoint configurations while safely staying within track bounds. We achieve zero-shot deployment on the car by training the agent in a simulation environment built on top of a neural stochastic differential equation vehicle model learned from pre-collected driving data. Experiments on a Toyota GR Supra and Lexus LC 500 show that the agent is capable of drifting smoothly through varying waypoint configurations with tracking error as low as 10 cm while stably pushing the vehicles to sideslip angles of up to 63°.

Model predictive control (MPC) algorithms can be sensitive to model mismatch when used in challenging nonlinear control tasks. In particular, the performance of MPC for vehicle control at the limits of handling suffers when the underlying model overestimates the vehicle's performance capabilities. In this work, we propose a risk-averse MPC framework that explicitly accounts for uncertainty over friction limits and tire parameters. Our approach leverages a sample-based approximation of an optimal control problem with a conditional value at risk (CVaR) constraint. This sample-based formulation enables planning with a set of expressive vehicle dynamics models using different tire parameters. Moreover, this formulation enables efficient numerical resolution via sequential quadratic programming and GPU parallelization. Experiments on a Lexus LC 500 show that risk-averse MPC unlocks reliable performance, while a deterministic baseline that plans using a single dynamics model may lose control of the vehicle in adverse road conditions.

Vehicle mobility and control performance on deformable terrains is governed by the complex interaction that occurs at the tire-terrain interface. Unfortunately, on deformable terrains, accurately measuring terrain information is challenging, and discrepancies between assumed and actual parameters can degrade control performance and cause a loss of vehicle mobility. To address these challenges, this paper proposes an online adaptive Model Predictive Control (MPC) framework for autonomous vehicles operating in off-road environments with deformable terrains. First, we develop a physics-informed learning tire model for deformable terrains that is adaptable online and compatible with MPC. A novel Model Predictive Control formulation is presented for autonomous vehicles operating on deformable terrains and the efficacy of the formulation and proposed tire model is evaluated in simulation with Project Chrono. Comparative experiments, with and without online adaptation, highlight improved speed and path tracking performance through online adaptation when a mismatch between assumed and actual terrain parameters is present.

In this paper, a control barrier function (CBF) is constructed for the lane-keeping problem which is applicable to both human-driven and automated vehicles. Based on the resulting CBF, a safety filter is developed that prevents the vehicle from crossing the lane boundaries, while only modifying the nominal steering input when necessary. The effectiveness of the proposed control approach is demonstrated in a series of numerical simulations and real vehicle experiments with a human driver. The experimental results show that the safety filter can successfully keep the vehicle inside the lane boundaries by seamlessly modifying the steering input of the human driver in a minimally invasive manner.

Machine learning models play a key role in safety-critical applications, such as autonomous vehicles and advanced driver assistance systems, where their robustness during inference is essential to ensure reliable operation. Sensor faults, however, can corrupt input signals, potentially leading to severe model failures that compromise reliability. In this context, pretraining emerges as a powerful approach for learning expressive representations applicable to various downstream tasks. Among existing techniques, masking represents a promising direction for learning representations that are robust to corrupted input data. In this work, we extend this concept by specifically targeting robustness to sensor outages during pretraining. We propose a self-supervised masking scheme that simulates common sensor failures and explicitly trains the model to recover the original signal. We demonstrate that the resulting representations significantly improve the robustness of predictions to seen and unseen sensor failures on a vehicle dynamics dataset, maintaining strong downstream performance under both nominal and various fault conditions. As a practical application, we deploy the method on a modified Lexus LC 500 and show that the pretrained model successfully operates as a substitute for a physical sensor in a closed-loop control system. In this autonomous racing application, a supervised baseline trained without sensor failures may cause the vehicle to leave the track. In contrast, a model trained using the proposed masking scheme enables reliable racing performance in the presence of sensor failures.

Drifting, a skillful driving technique involving intentional traction loss and counter-steering, traditionally demands high-speed maneuvers under high-friction conditions, posing significant risks and fear for novices. Our study explores low-µ (low friction) emulation, simulating icy conditions to facilitate drift training at safer, lower speeds. This approach not only enhances safety and mitigates fear by reducing the required speed for drifting, but also extends the time for them to react. A between-group design was employed, comparing drift training outcomes between participants trained exclusively in higher-µ conditions (control group) and those who trained initially in lower-µ conditions before transitioning to higher-µ conditions (target group). The performance was assessed through the average distance of continuous sliding, along with subjective measures of motivation and workload. The results showed that the target group achieved greater slide distances in the retention session and reported higher scores on the positive intrinsic motivation factors, suggesting enhanced performance and engagement.

Autonomous vehicle controllers capable of drifting can improve safety in dynamic emergency situations. However, drifting involves operating at high sideslip angles, which is a fundamentally unstable operating regime that typically requires an accurate vehicle model for reliable operation; such models may not be available after environmental or vehicle parameter changes. Towards that goal, this work presents a Nonlinear Model Predictive Control approach which is capable of initiating and controlling a drift in a production vehicle even when changes in vehicle parameters degrade the original model. A neural network model of the vehicle dynamics is used inside the optimization routine and updated with online learning techniques, giving a higher fidelity and more adaptable model. Experimental validation on a full size, nearly unmodified Lexus LC500 demonstrates the increased modeling fidelity, adaptability, and utility of the presented controller framework. As the LC500 is a difficult car to drift, previous approaches which rely on physics based vehicle models could not complete the autonomous drift tests on this vehicle. Furthermore, the tires on the experimental vehicle are then switched, changing the vehicle parameters, and the capability of the controller to adapt online is demonstrated.