Collision Avoidance with Stochastic Model Predictive Control for Systems with a Twofold Uncertainty Structure

Brüdigam, Tim; Zhan, Jie; Wollherr, Dirk; Leibold, Marion · 2021 · OpenAlex-citations

DOI: 10.1109/itsc48978.2021.9564589

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of collision avoidance for automated systems, particularly vehicles, operating in environments with a "twofold" uncertainty structure. The authors identify that predicting the behavior of surrounding agents involves two distinct types of uncertainty: high-level task uncertainty (e.g., whether an agent will change lanes or keep its lane) and low-level task execution uncertainty (e.g., the specific trajectory and timing of that maneuver). Existing Model Predictive Control (MPC) methods often handle only one type of uncertainty or are overly conservative. To resolve this, the paper proposes a generalized Stochastic Model Predictive Control (S+SC MPC) framework that efficiently combines Scenario MPC (SCMPC) for task uncertainty and analytic SMPC for execution uncertainty. The methodology formulates an optimal control problem where the controlled agent plans its trajectory while avoiding dynamic obstacles. Task uncertainty is handled by drawing a limited number of samples from the probability distribution of possible tasks, ensuring that even the least likely tasks are accounted for within a predefined risk tolerance. Task execution uncertainty, modeled as additive Gaussian noise, is handled using an analytic approximation of chance constraints, which linearizes the safety requirements around nominal trajectories. This combination allows the controller to consider multiple potential future behaviors of surrounding agents without the high computational cost of sampling-based methods for Gaussian noise or the conservatism of robust control methods. The framework is designed to be applicable to multiple obstacles, each with multiple possible tasks. The proposed S+SC MPC framework is evaluated through a simulation study involving an ego vehicle navigating a highway scenario with five target vehicles. The simulation models the target vehicles as having nine possible maneuvers, including lane changes, acceleration, and braking, each with associated probabilities. The results demonstrate that the S+SC MPC approach successfully avoids collisions while maintaining efficient motion. Compared to stand-alone SMPC or SCMPC methods, the combined framework effectively manages the mixed uncertainty structure. The study shows that the controller can adapt to varying scenario configurations and maneuver probabilities, providing a tractable solution for real-time collision avoidance in complex, uncertain environments. The significance of this work lies in its generalization of previous S+SC MPC approaches, extending them from simple single-vehicle scenarios to complex environments with multiple agents and tasks. By efficiently handling both discrete task choices and continuous execution variations, the proposed method offers a practical solution for automated driving systems. It balances safety and efficiency by allowing a small, controlled probability of constraint violation rather than enforcing hard constraints that lead to conservative behavior. This contributes to the field of intelligent transportation systems by providing a robust control strategy that can anticipate and react to the diverse and uncertain behaviors of surrounding traffic.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• anticipation