Controller Adaptation via Learning Solutions of Contextual Bayesian Optimization

(Experiment Results)

Authors: Viet-Anh Le and Andreas A. Malikopoulos

Summary

This is the complementary website for showing the experiment results in the paper. This paper proposes a framework to address the controller adaptation problem for dynamical systems with task-dependent or time-varying parameters via learning the solutions of contextual Bayesian optimization with Gaussian processes. We demonstrate the practicality of the framework through a sim-to-real application, where the weighting strategy of model predictive control for connected and automated vehicles interacting with human-driven vehicles is learned from simulations and applied in real-time experiments. The paper preprint can be found here.

Experiments with a single HDV

We consider an intersection scenario in a robotic testbed called the Information and Decision Science Lab Scaled Smart City (IDS3C). In our experimental setup, a robotic car is manually controlled by a human participant using a driving emulator to generate realistic human-driven vehicle behavior. We show four specific simulations, each demonstrating different driving styles generated by the human participant. In the video, the black vehicle is a CAV, while the blue vehicle is an HDV.

Experiments with two HDVs

We test the controller in a scenario involving two HDVs. In the video, the yellow vehicle is a CAV, while the black and blue vehicles are HDVs.

Page updated

Report abuse