Safety-Aware Pursuit-Evasion Games in Unknown Environments Using Gaussian Processes and Finite-Time Convergent...

Area of Science:

• Robotics and Control Systems
• Artificial Intelligence
• Game Theory

Background:

• Pursuit-evasion games are critical for autonomous systems.
• Adapting to unknown, cluttered environments poses significant challenges.
• Ensuring safety during dynamic interactions is paramount.

Purpose of the Study:

• To develop a safe pursuit-evasion game enabling finite-time capture.
• To achieve optimal performance in unknown, cluttered environments.
• To enable adaptive pursuit-evasion policies using reinforcement learning.

Main Methods:

• Formulation as a zero-sum differential game.
• A critic-only reinforcement learning (RL) algorithm for policy learning.
• Integration of barrier functions for safety in cluttered environments.
• Gaussian processes (GPs) for adaptive environment learning.

Main Results:

• Finite-time capture of the evader is achieved.
• Safe navigation and adaptation to unknown environments demonstrated.
• Optimal pursuit-evasion policies were learned online.
• Simulation results validate the proposed approach's efficacy.

Conclusions:

• The proposed RL-based safe pursuit-evasion game effectively achieves finite-time capture.
• The integration of barrier functions and GPs ensures safety and adaptability.
• This approach offers a robust solution for autonomous systems operating in complex environments.