Decision-making in automated driving is influenced both by objective traffic rules and subjective perceptions and goals of the driver. Thus, a suitable representation of the environment of the autonomous vehicle is required to model complex traffic situations and extract key features. To achieve this objective, this work uses an ontology-based situation interpretation (OBSI) to model traffic situations. The resulting semantic state representation is used to train models of vehicle-controlling agents using reinforcement learning. Based on our simulations, it can be shown that the semantic preprocessing of traffic situations significantly improves the agent's performance regarding safety and driving style.
A Hybrid Tactical Decision-Making Approach in Automated Driving Combining Knowledge-Based Systems and Reinforcement Learning
2022-10-08
687435 byte
Conference paper
Electronic Resource
English
TACTICAL DECISION-MAKING IN AUTONOMOUS DRIVING BY REINFORCEMENT LEARNING WITH UNCERTAINTY ESTIMATION
British Library Conference Proceedings | 2020
|UNCERTAINTY-DIRECTED TRAINING OF A REINFORCEMENT LEARNING AGENT FOR TACTICAL DECISION-MAKING
European Patent Office | 2023
|