· 1 min read
MSc. Computer Science (Robotics) - Research
MSc Research: Multi-Agent Reinforcement Learning
My Master of Science in Computer Science (Robotics) at the University of the Witwatersrand focuses on Multi-Agent Reinforcement Learning (MARL) within the RAIL Lab. My research addresses the challenge of communication efficiency in decentralized multi-robot systems.
Problem Statement
In real-world deployments (e.g., search and rescue, underground mining), communication bandwidth is often limited or unreliable. Standard MARL approaches often assume perfect, unlimited communication, which leads to failure in these constrained environments.
Methodology
- Communication Minimization: Developing a framework where agents learn a minimal communication protocol to coordinate their actions.
- Decentralized Execution: Agents act based on local observations and learned communication policies, making the system robust to node failures.
- Environments: Testing in complex, dynamic grid-worlds and continuous control tasks using vectorized gym-like environments.
Goals
To contribute a robust, bandwidth-aware MARL algorithm that can be deployed on physical UAV swarms operating in GPS-denied and communication-constrained environments.
