Development of learning-based multi-agent models for validating automated driving functions in adaptive traffic simulations

Context

Validation of automated driving functions requires realistic traffic simulations to reliably test behavior in complex scenarios. Traditional simulation approaches based on pre-recorded trajectories are limited because they cannot respond to the behavior of automated vehicles. This leads to inaccurate results and makes it difficult to evaluate driving systems in dynamic traffic situations.

Multi-Agent Reinforcement Learning (MARL) is a promising tool to address this problem by enabling the development of intelligent, adaptive traffic agents. These agents can interact, learn, and adapt their strategies in real time to the environment and the behavior of automated vehicles. With these reactive and adaptive models, critical traffic scenarios can be dynamically generated, allowing for more accurate and robust validation of driving functions under realistic conditions.

Goals

• Conceptualization and implementation of a multi-agent system for simulating traffic agents
• Use of reinforcement learning to develop adaptive traffic agents that flexibly adjust to different traffic situations
• Investigation and development of approaches for simulating critical scenarios for automated vehicles
• Validation of the system through simulation and analysis of traffic situations

Requirements

• Interest in the topics of multi-agent systems, reinforcement learning, and traffic simulation
• Programming skills in Python, ideally with experience in ML libraries (e.g., PyTorch)
• Motivation and creative thinking