About
Solving OpenAI's Bipedal Walker environment using Proximal Policy Optimization (PPO) algorithm. For comparison an implementation with DDPG algorithm is also provided.
Results
Project Structure
2 different solutions have been implemented in this repo
ppo
: directory for ppo agentppo_continuous.py
: code for running the ppo agent for the bipedal environmentppo_bipedal.ipynb
: jupyter notebook for agent training and visualisationutils.py
: utility functionsdeep_rl
: directory with modular functions for the PPO agent
ddpg
: directory for ddpg agentDDPG.ipynb
: jupyter notebook for training the agentddpg_agent.py
: code for the agent model, experience replay and OU noisemodel.py
: actor and critic networks
Instructions
Open ppo/ppo_bipedal.ipynb
to see an implementation of PPO with OpenAI Gym's BipedalWalker environment.