An AI agent learning to walk in gym's BipedalWalker environment.

This AI agent uses Evolution Strategies and deep learning models to learn how to walk.

Read Evolution Strategies as a Scalable Alternative to Reinforcement Learning from OpenAI if you are interested.

After a few hundred iterations, he can walk! Or he may run ;)

Dependencies

• evostra
• gym

Usage

To see the agent walking:

from bipedal import *

agent = Agent()

# the pre-trained weights are saved into 'weights.pkl' which you can use.
agent.load('weights.pkl')

# play one episode
agent.play(1)

To start training the agent:

# train for 100 iterations
agent.train(100)