Learning-Based Safe Motion Control of Vehicle Ski-Stunt Maneuvers

TL;DR

This paper introduces a learning-based control approach for autonomous vehicle ski-stunt maneuvers, combining Gaussian process models with control barrier functions to ensure safety and balance during complex maneuvers.

Contribution

It develops a probabilistic control framework that integrates Gaussian process modeling with safety guarantees for vehicle ski-stunt maneuvers.

Findings

Successfully avoids obstacle collisions.

Maintains vehicle balance during maneuvers.

Validated through simulations and scaled experiments.

Abstract

This paper presents a safety guaranteed control method for an autonomous vehicle ski-stunt maneuver, that is, a vehicle moving with two one-side wheels. To capture the vehicle dynamics precisely, a Gaussian process model is used as additional correction to the nominal model that is obtained from physical principles. We construct a probabilistic control barrier function (CBF) to guarantee the planar motion safety. The CBF and the balance equilibrium manifold are enforced as the constraints into a safety critical control form. Under the proposed control method, the vehicle avoids the obstacle collision and safely maintain the balance for autonomous ski-stunt maneuvers. We conduct numerical simulation validation to demonstrate the control design. Preliminary experiment results are also presented to confirm the learning-based motion control using a scaled RC truck for autonomous ski-stunt maneuvers.