The field of autonomous vehicle research is moving towards more sophisticated and dynamic modeling of complex interactions between multiple agents. This is evident in the development of novel architectures and frameworks that can capture the evolving nature of these interactions and improve prediction accuracy. One of the key areas of focus is on joint multi-agent motion forecasting, which is crucial for safe planning and coordination of autonomous vehicles. Researchers are also exploring the use of reinforcement learning and digital twin-based approaches to enhance safety and efficiency in various traffic scenarios. Noteworthy papers in this area include:

• ProgD, which achieves state-of-the-art performance on multi-agent prediction benchmarks through progressive multi-scale decoding with dynamic graphs.
• Platoon-Centric Green Light Optimal Speed Advisory, which develops a safe RL-based system for optimizing CAV speed while ensuring car-following and red-light safety.
• STEP, which introduces a new benchmarking framework for trajectory prediction models that supports heterogeneous traffic scenarios and joint prediction models.
• Digital Twin-based Cooperative Autonomous Driving, which proposes a DT-based cooperative driving system with a hybrid RL framework for enhancing safety and efficiency at unsignalized intersections.