Research Objectives

Development of a robust navigation system that enables autonomous robots to navigate complex, dynamic environments while coordinating with multiple agents and avoiding both static and moving obstacles.

Methodological Approach

• Deep Q-Networks (DQN) with prioritized experience replay
• Multi-agent coordination using centralized training, decentralized execution
• Sim-to-real transfer with domain randomization techniques
• Safety constraints implemented through constrained policy optimization

Experimental Setup

• Simulation environment: Custom-built Unity3D scenarios
• Physical testing: TurtleBot3 robots in laboratory settings
• Evaluation metrics: Success rate, path efficiency, collision avoidance

Key Achievements

• 89% success rate in complex multi-agent scenarios
• 40% improvement in path efficiency compared to traditional planners
• Successful sim-to-real transfer with minimal performance degradation
• Real-time decision making with 10Hz control frequency