The recent developments in the research area of autonomous systems and multi-agent interactions have shown a significant shift towards enhancing safety, efficiency, and robustness through advanced learning and control strategies. A common thread among the innovative works is the integration of game theory and reinforcement learning to address complex, multi-agent scenarios. This approach allows for the development of adaptive, decentralized solutions that can handle dynamic environments and interactions between agents. Notably, there is a strong emphasis on ensuring safety through state-wise constraints and controlled invariant sets, which provide a robust framework for maintaining safety in real-time applications. Additionally, the use of multi-objective reinforcement learning is gaining traction for addressing the conflicting goals in autonomous systems, such as balancing defensive actions with maintaining network functionality. The field is also witnessing advancements in scenario generation and testing, where adversarial learning frameworks are being employed to stress-test autonomous systems and iteratively harden their safety mechanisms. Overall, the research is moving towards more sophisticated, adaptive, and resilient systems that can operate effectively in complex, real-world conditions.
Enhancing Safety and Efficiency in Autonomous Systems through Advanced Learning and Control
Sources
Energy Efficient Automated Driving as a GNEP: Vehicle-in-the-loop Experiments
A Systematic Study of Multi-Agent Deep Reinforcement Learning for Safe and Robust Autonomous Highway Ramp Entry
Path Planning and Task Assignment for Data Retrieval from Wireless Sensor Nodes Relying on Game-Theoretic Learning
Safe Multi-Agent Reinforcement Learning with Convergence to Generalized Nash Equilibrium
On Multi-Agent Inverse Reinforcement Learning
Avoiding Deadlocks Is Not Enough: Analysis and Resolution of Blocked Airplanes
CRASH: Challenging Reinforcement-Learning Based Adversarial Scenarios For Safety Hardening
Multi-Objective Reinforcement Learning for Automated Resilient Cyber Defence
Rate-Informed Discovery via Bayesian Adaptive Multifidelity Sampling