A Tutorial on Gaussian Process Learning-based Model Predictive Control
Organizations: Concordia University, 1455 Blvd. De Maisonneuve Ouest, Montreal, QC, Canada, H3G 1M8
Abstract
This tutorial provides an accessible and implementation-oriented introduction to Gaussian process learning-based model predictive control (GP-MPC), which combines probabilistic residual modeling with receding-horizon control. Its central tutorial contribution is a detailed, step-by-step derivation of multi-step mean and covariance propagation for GP-augmented prediction models. The derivation shows how commonly used propagation formulas follow from the laws of total expectation and total covariance, and clarifies the roles of the GP posterior mean, GP posterior covariance, uncertain inputs, and query-output cross-covariances. Building on this foundation, the article distinguishes mean-only unconstrained GP predictive control from uncertainty-aware constrained GP-MPC, clarifies regulation and output-tracking formulations, and provides concise implementation and computational-complexity guidance, including practical software-tool references and discussion of the scope of closed-loop guarantees. Mobile-robot path-following examples illustrate mean-only unconstrained implementations, whereas mixed-vehicle platooning illustrates uncertainty-aware constrained GP-MPC with uncertainty-dependent safety-distance tightening. The tutorial is intended to help researchers and practitioners understand, implement, and critically evaluate GP-MPC designs for robotic and autonomous systems operating under modeling uncertainty.
Figures & tables
| Method | GP mean | GP variance | Uncertain-input contribution | Role in this tutorial |
|---|---|---|---|---|
| Mean-only prediction | Yes | No | No covariance propagation | Path-following examples |
| Mean-equivalent covariance | Yes | Yes | GP cross-covariance neglected | Mixed-platoon example |
| Taylor covariance | Yes | Yes | First-order sensitivity through | Main step-by-step derivation |
| Moment matching | Yes | Yes | Moment integration under Gaussian input | Alternative discussed |
| Application | GP information used online | Constraints in the presented implementation | Formulation classification |
|---|---|---|---|
| GP-NMPC path following | Posterior mean | No explicit inequality constraints | Mean-only unconstrained GP predictive control |
| GP-FBLMPC path following | Posterior mean | No explicit inequality constraints | Mean-only unconstrained GP predictive control |
| Mixed-platoon GP-MPC | Mean and propagated variance | Spacing, speed, acceleration, and chance constraints | Uncertainty-aware constrained GP-MPC |
| Dynamic sparse GP-MPC | Sparse mean and variance | Retains the mixed-platoon constraints | Sparse constrained GP-MPC |
Model Predictive Control
Model Predictive Control optimizes future actions with a system model, repeatedly applying the first action as new information arrives.
See all 336 papers