Open Collections

Ren, Kai

University of British Columbia

In this work, we present a framework for trajectory planning of autonomous vehicles in the presence of multimodal environmental uncertainty. The trajectory planning framework is designed based on chance-constrained and Conditional Value-at-risk (CVaR)-constrained programming methods. We employ a Gaussian Mixture Model (GMM) to model the multimodal behaviors of the uncertain future states of the dynamic obstacles. We first summarize tractable chance-constrained and CVaR-constrained optimization techniques under GMM uncertain parameters. We consider the case in which the GMM moments are known and the data-driven scenario when only samples are provided. We then formulate both the chance-constrained and CVaR-constrained finite horizon trajectory planning (TP) problem, with a time-varying deterministic linear dynamic system and polyhedral obstacles showing GMM uncertainty. We present tractable reformulations of both chance-constrained and CVaR-constrained TP problems in cases of known GMM moments and when only samples of the uncertainty are available. When the GMM moments are estimated via finite samples, we present a tight concentration bound to ensure the satisfaction of the chance constraint with a probabilistic guarantee. Finally, we present a closed-loop Model Predictive Control algorithm under GMM uncertainty, which employs periodically updated observations and predictions of the uncertain obstacles, and solves the TP problems in a receding horizon fashion. We verify our methods by conducting numerical simulations, integrating with state-of-the-art trajectory prediction algorithms, and testing on real-world autonomous driving datasets.

Text

2022-10-19

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/82910

Electrical and Computer Engineering

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/