CCAM-DI: Reinforcement Learning-Based Autonomous Intersection Control

This dataset contains the code, configurations, and trained models for a reinforcement learning (RL)-based autonomous intersection control system developed using a multi-agent framework and the SUMO traffic simulator.

The first study (On Performance Improvement of Reinforcement Learning for Collision Avoidance in Autonomous Intersections) introduces methods to improve RL training efficiency by filtering irrelevant observations and actions, leading to better performance in collision avoidance.

The second study (Evaluating the Robustness of RL-based Autonomous Intersection Control to Data Integrity Attacks) extends this work by analyzing the robustness of the trained policies under data integrity attacks, where malicious agents manipulate position information.

The dataset includes implementation files, environment settings, and trained models that allow users to reproduce the experiments and evaluate both performance improvements and robustness aspects. Users can run the provided scripts to train or test policies under normal and adversarial conditions, making the dataset suitable for benchmarking and further research on secure autonomous traffic control.