Application of Responsibility-Sensitive Safety in Areas with Limited Visibility: Occlusions in RSS
Published in ‘2023 IEEE 26th International Conference on Intelligent Transportation Systems’, IEEE 2023
Abstract
Areas of limited visibility are common in day-to-day traffic: be it static buildings, parked vehicles, traffic participants blocking the vehicle’s line of sight, harsh weather conditions or just narrow curves that impede the automated driving sensor suite to inspect the road ahead. Autonomous vehicles have to be able to safely cope with this kind of constraints. The Responsibility-Sensitive Safety model (RSS) demands vehicles to exercise caution with respect to occlusions and to consider also occluded road agents. This paper provides a concrete implementation of how occlusions in RSS can be addressed and investigates the balance between safety and usefulness of the model when a reasonably foreseeable behavior of occluded road agents is assumed. We perform occlusion experiments in urban as well as on highway scenarios with the driving simulation platform CARLA applying different parameterization of the agents kinematic properties and the safety model parameters to analyse and judge the consequences with respect to safe driving and overcautious driving behaviors.
Key Contributions
- Safety analysis and validation
- Implements occlusion handling with artificial vehicles
- Evaluates approach in urban/highway scenarios
- Shows aggressive parameters can reduce usability
- Analyzes real-world data for assumptions
- Demonstrates collision mitigation via simulation