Welcome to the exploration of bipedal locomotion mastery in the Bipedal-Walking-v3 environment! This project encompasses the application of a modified Soft Actor-Critic (SAC) algorithm as a part of my Reinforcement Learning course, aiming to skillfully navigate through the challenges and complexities of bipedal movement.
"This paper implements Soft Actor-Critic(SAC) in both standard and hardcore versions of BipedalWalker-v3 by adjusting the learning rate and reward scale. Both versions achieved over 300 points at episodes 117 and 1296 respectively. While the standard version converged within 200 episodes, it was difficult to determine if the hardcore version converged due to frequent crashes of the NVIDIA CUDA Center (NCC) and limited time. Future work should focus on running the agent in a more stable environment and exploring adaptive learning rate methods to optimize performance in complex environments."
Achieved a noteworthy score of 300 at episode 117, demonstrating a promising convergence in the initial 200 episodes.
Successfully garnered a reward of 300 at episode 1296, showcasing adept capabilities in navigating through a notably challenging scenario.
Ensure the availability of the following before proceeding:
- Python 3.x
- OpenAI Gym
- PyTorch
Execute the following commands in your terminal to set up and run the project:
git clone [repository_url]
cd SAC-with-Auto-tuned-Temperature-for-Bipedal-Walking-v3