Bohn, Christopher (Karlsruhe Institute of Technology (KIT)), Bosch, Janne (KIT), Hess, Manuel (KIT), Hohmann, Soeren (Karlsruhe Institute of Technology)

Reducing Conservatism in Fast and Safe Motion Generation by Means of Captivity-Escape Games

Scheduled for presentation during the Invited Session "S17b-Synthetic-Data-Aided Safety-Critical Scenario Understanding in ITS" (TH-EA-T17), Thursday, November 20, 2025, 14:10−14:30, Southport 2

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 18, 2025

Abstract

This paper addresses conservatism in existing methods for ensuring safety in online model-based motion generation, known as fast planning and safe tracking. Computational constraints restrict online motion planning to low-fidelity planning models. Crucially, planning with low-fidelity models risks safety, as the dynamic feasibility of resulting reference trajectories is not ensured, potentially causing inevitable tracking errors that may result in collisions with obstacles. Existing methods address this safety risk by augmenting obstacles with a safety margin that prevents collisions under worst-case tracking errors. However, these methods only utilize a single or a limited set of safety margins, which often results in overly conservative reference trajectories, as either the applied safety margin is excessively large, or the planning model's performance is overly conservative. To address this conservatism, we present a framework that leverages a recently presented method for online adaptation of the planning model's performance to a given safety margin. We demonstrate the presented framework in a real-world experiment, showing that it can online adapt the planning model's performance to the largest safety margin that is admissible in the current planning environment.