Thrilling driving at the intersection of AI, machine learning, and automotive engineering
“Autonomous Drifting With 3 Minutes of Data Via Learned Tire Models,” by Franck Djeumou, Jonathan Y.M. Goh, Ufuk Topcu, and Avinash Balachandran from University of Texas at Austin, and Toyota Research Institute, in Los Altos, Calif.
Presented at the 2023 IEEE International Conference on Robotics and Automation (ICRA)
Picture this: a sleek Toyota Supra gracefully sliding sideways, tires screeching, as it effortlessly navigates a tight corner. What if we told you that achieving such automotive acrobatics could now be done autonomously? A groundbreaking research paper titled "Autonomous Drifting With 3 Minutes of Data Via Learned Tire Models" by Franck Djeumou, Jonathan Y.M. Goh, Ufuk Topcu, and Avinash Balachandran from the University of Texas at Austin and the Toyota Research Institute in Los Altos, Calif., unveils a revolutionary approach to autonomous drifting, offering an exhilarating blend of performance and safety.
At the heart of this breakthrough lies the accurate modeling of tire forces near the limits of adhesion, where the interactions between the tires and the road become nonlinear and highly intricate. Traditionally, such modeling has been a challenge, but advancements in artificial intelligence and machine learning have opened up new frontiers. By leveraging neural ordinary differential equations and a neural-ExpTanh parameterization, the researchers have developed a novel family of tire force models that not only meet physically insightful assumptions but also possess the fidelity necessary to capture higher-order effects directly from vehicle state measurements.
The significance of these tire force models cannot be overstated, as they offer a path to improve safety, especially in emergency situations where high forces are required. With the ability to accurately predict and control tire behavior, autonomous vehicles can navigate challenging maneuvers with precision, reducing the risk of accidents and enhancing passenger protection.
To put these models to the test, the researchers incorporated them into an existing nonlinear model predictive control framework, replacing the traditional analytical brush tire model. The results were nothing short of astounding. In a series of experiments with a customized Toyota Supra, the autonomous system demonstrated high-performance drifting capabilities on various trajectories, reaching speeds of up to 45 miles per hour. What's even more remarkable is that this level of performance was achieved with less than three minutes of driving data—a scarcity that underscores the efficiency and effectiveness of the proposed approach.
Comparisons with the benchmark model further validated the superiority of the neural tire force models. The new models showcased a fourfold improvement in tracking performance, delivering smoother control inputs and faster, more consistent computation times. These advancements not only pave the way for breathtaking autonomous drifting experiences but also hold immense potential for enhancing overall vehicle dynamics and control.
The researchers delved deeper into the fascinating research behind autonomous drifting with learned tire models and explored the foundations of the neural ordinary differential equations and neural-ExpTanh parameterization, unravel the experimental setup, and analyze the impressive results that mark a new era in autonomous vehicle capabilities. So buckle up, as we embark on a thrilling journey at the intersection of AI, machine learning, and automotive engineering. Get ready to witness the fusion of exhilaration and safety like never before.
Abstract: Near the limits of adhesion, the forces generated by a tire are nonlinear and intricately coupled. Efficient and accurate modeling in this region could improve safety, especially in emergency situations where high forces are required. To this end, we propose a novel family of tire force models based on neural ordinary differential equations and a neural-ExpTanh parameterization. These models are designed to satisfy physically insightful assumptions while also having sufficient fidelity to capture higher-order effects directly from vehicle state measurements. They are used as drop-in replacements for an analytical brush tire model in an existing nonlinear model predictive control framework. Experiments with a customized Toyota Supra show that scarce amounts of driving data—less than 3 minutes—is sufficient to achieve high-performance autonomous drifting on various trajectories with speeds up to 45 miles per hour. Comparisons with the benchmark model show a 4x improvement in tracking performance, smoother control inputs, and faster and more consistent computation time.
