Battery-Constrained Single-Agent Exploration
Project Overview
This project focuses on intelligent exploration in resource-constrained environments using a battery-aware single-agent. Built on a modified WindyGridWorld environment, the agent learns to maximize coverage while avoiding energy depletion using DDPG.
Technologies Used
- Python (Gym-like custom environments)
- Reinforcement Learning: DDPG
- Battery-Constrained Planning
- Matplotlib for visualization
Key Features
- Continuous state and action space
- Reward shaping for battery consumption
- Dynamic environment with stochasticity
- Custom visualization of coverage and battery use
Research Contributions
- Novel reward formulation incorporating battery limits
- Evaluation of DDPG performance in sparse reward exploration tasks
- Potential extension to multi-agent systems