rl_bicycle

This environment features two reinforcement learning agents designed to learn bicycle riding, each using a different algorithm. The first agent employs SARSA(λ), an on-policy algorithm that balances exploration and exploitation by updating Q-values based on the current state-action pair and future rewards. SARSA(λ) improves the agent’s performance through a combination of learning from past experiences and adjusting its strategy over time.

The second agent uses Proximal Policy Optimization (PPO), a stable and robust algorithm particularly suited for continuous action spaces. PPO optimizes a surrogate objective function to improve the agent’s policy while preventing drastic updates, ensuring more stable learning. This setup provides a valuable comparison between the two algorithms, showcasing their unique approaches to agent learning and performance improvement in dynamic environments.