Real-Time Online Learning for Model Predictive Control using a Spatio-Temporal Gaussian Process Approximation
This work addresses a computational bottleneck for real-time GP-MPC applications, offering an incremental improvement for control systems like autonomous racing.
The paper tackles the high computational cost of online learning in Gaussian process-based model predictive control (GP-MPC) by proposing an efficient spatio-temporal GP approximation with constant complexity, enabling real-time learning and improved control performance, as demonstrated in simulations and hardware experiments for autonomous miniature racing.
Learning-based model predictive control (MPC) can enhance control performance by correcting for model inaccuracies, enabling more precise state trajectory predictions than traditional MPC. A common approach is to model unknown residual dynamics as a Gaussian process (GP), which leverages data and also provides an estimate of the associated uncertainty. However, the high computational cost of online learning poses a major challenge for real-time GP-MPC applications. This work presents an efficient implementation of an approximate spatio-temporal GP model, offering online learning at constant computational complexity. It is optimized for GP-MPC, where it enables improved control performance by learning more accurate system dynamics online in real-time, even for time-varying systems. The performance of the proposed method is demonstrated by simulations and hardware experiments in the exemplary application of autonomous miniature racing.