Reinforcement Learning Algorithm Implementations on MuJOCO

• Tech Stack: Python, PyTorch, MuJoCo
• Github URL: Project Link

Implemented Soft Actor-Critic (SAC), Proximal Policy Optimization (PPO), and model-based reinforcement learning algorithms in PyTorch from scratch. Trained agents in various MuJoCo environments, including complex continuous control tasks such as spider, walker, and humanoid locomotion. Each algorithm was customized to handle the dynamic and continuous nature of these environments, with SAC focusing on efficient exploration, PPO ensuring stability in policy updates, and model-based RL leveraging environment models for optimized learning.

Conducted a comparative study and analysis to evaluate the performance of each algorithm across different control tasks, analyzing factors like training speed, stability, and overall agent performance. The study highlighted the trade-offs between exploration, policy optimization, and model-based strategies, offering insights into the effectiveness of each approach for complex continuous control problems in simulated environments.