Data-Driven Model Identification via Hyperparameter Optimization for the Autonomous Racing System
Authors:
- Hyunki SeongHSHyunki Seong
- Chanyoung ChungCCChanyoung Chung
- David Hyunchul ShimDSDavid Hyunchul Shim
Preprint
Abstract
In this letter, we propose a model identification method via hyperparameter optimization (MIHO). Our method is able to identify the parameters of the parametric models in a data-driven manner. We utilize MIHO for the dynamics parameters of the AV-21, the full-scaled autonomous race vehicle, and integrate them into our model-based planning and control systems. In experiments, the models with the optimized parameters demonstrate the generalization ability of the vehicle dynamics model. We further conduct extensive field tests to validate our model-based system. The tests show that our race systems leverage the learned model dynamics well and successfully perform obstacle avoidance and high-speed driving over $200 km/h$ at the Indianapolis Motor Speedway and Las Vegas Motor Speedway. The source code for MIHO and videos of the tests are available at https://github.com/hynkis/MIHO.
Keywords
All Related Versions
This publication has 5 references indexed in Scilit:
- AMZ Driverless: The full autonomous racing systemJournal of Field Robotics, 2020
- End-to-End Neural Network for Vehicle Dynamics Modeling2020
- Neural network vehicle models for high-performance automated drivingScience Robotics, 2019
- Race Car Design2014
- Tyre Modelling for Use in Vehicle Dynamics StudiesSAE International Journal of Advances and Current Practices in Mobility, 1987