Reproduce “Learning Model Predictive Control with Error Dynamics Regression for Autonomous Racing” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Haoru Xue, Edward L. Zhu, John M. Dolan, Francesco Borrelli
A Learning Model Predictive Control strategy employs error dynamics regression, learning the difference between a nominal physics-based model and true vehicle dynamics. This approach demonstrated increased robustness to data scarcity and parameter tuning in simulations, reduced failures on a 1/10th scale autonomous car, and enabled iterative performance improvement on a full-size Indy Autonomous Challenge race car.
Key insights
- Learning the error dynamics provides greater robustness than directly regressing the full system dynamics, especially with sparse data or non-optimally tuned parameters.
- A well-designed regularization in the error regression allows the controller to default to a known, stable nominal model when insufficient local data is available.
- Combining a global, physics-based model with local, data-driven error correction enables safer exploration and iterative performance improvement at the limits of vehicle handling.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.