This repository contains implementations of various reinforcement learning algorithms (DQN, DDQN TD3, PPO) for solving the Lunar Lander-v2 and BipedalWalker-v3 environment from OpenAI's Gym.

• DQN (Deep Q-Network)
• DDQN (Double Deep Q-Network)
• TD3 (Twin Delayed Deep Deterministic Policy Gradient)
• PPO (Proximal Policy Optimization)

git clone https://github.com/Viswesh-N/RL_BipedWalker.git
cd RL_BipedWalker

It’s recommended to use a virtual environment for managing dependencies.

python -m venv env
source env/bin/activate  # On Windows use `env\Scripts\activate`

Before running any commands, ensure you install all required Python packages.

pip install -r requirements.txt

To ensure that imports aren't broken, run the following command in the root directory of the repository:

export PYTHONPATH=$(pwd)

Train the DQN or DDQN agent on the LunarLander-v2 environment:

python scripts/run.py --algo dqn --config_file experiments/dqn/dqn_config.yaml --seed 1 --eval_interval 10000 --num_eval_trajectories 10 --num_render_trajectories 0 --log_interval 1000

To run the DDQN agent, simply use the ddqn_config.yaml file in place of the dqn_config.yaml file. No other change is required.

Train the TD3 agent on the BipedalWalker-v3 environment:

python scripts/run.py --algo td3 --config_file experiments/td3/td3_config.yaml --seed 1 --eval_interval 10000 --num_eval_trajectories 10 --num_render_trajectories 0 --log_interval 1000

Train the PPO agent on the BipedalWalker-v3 environment:

python scripts/run.py --algo ppo --config_file experiments/ppo/ppo_config.yaml --seed 1 --eval_interval 10000 --num_eval_trajectories 10 --num_render_trajectories 0 --log_interval 1000

Training logs are stored in the data/ directory, organized by algorithm name, environment, and timestamp. You can visualize the training progress using TensorBoard:

tensorboard --logdir data/

Open https://localhost:6006 in your browser to view the logs.

You can view the plots for DQN and TD3 training in the images/ directory.

• DQN Plot:
• TD3 Plot:

You can modify the configurations in the experiments/ folder to change hyperparameters, learning rates, or other settings specific to each algorithm.

• LiDAR Raycasting Error: Sometimes, during training, you might encounter a r.LengthSquared() > 0.0f error due to numerical instability in the physics engine. The environment automatically resets when this occurs.