This repo provides robot learning environments for course Computational Intelligence and Robotics directed by Prof. Huaping Liu at Tsinghua University.

The task is training a (Bipedalwalker / Ant) to learn control and evolve its morph for better control.

Two difficulties of tasks are provided:

1. easy: training a (Bipedalwalker / Ant) on flat.
2. hard: training a (Bipedalwalker / Ant) on rough terrain.

cd docker
docker build -t 'evo-learning:v1' . 

Get the image_id.

docker run -it --name evo-learning -v ws:/root/ws -v /tmp/.x11-unix:/tmp/.x11-unix -e DISPLAY=$DISPLAY -e GDK_SCALE -e GDK_DPI_SCALE -p 9022:22 {image_id} /bin/bash

Important: The 'ws' is the docker volume where you should clone and put this repo, so that you can edit code inside the container.

Note: Replace the {image_id} with the image_id in the last step.

Entering containner using ssh with default passwd '123123':

ssh -Y [email protected] -p 9022

Note: It's necessary to use "-Y" option to allow X11 forwarding.

We provide a gym like environment api:

    env = make_env(args.env, render_mode="human")
    env.seed(seed=42)

    num_param = 8
    augment_vector = (1.0 + (np.random.rand(num_param)*2-1.0)*0.5)
    env.augment_env(augment_vector)
    observation, info = env.reset()
    for _ in range(1000):
        observation, reward, done, info = env.step(env.action_space.sample())
        env.render()
        if done:
            observation, info = env.reset()
    env.close()

ENVNAME can be one of:

AugmentBipedalWalker
AugmentBipedalWalkerSmallLegs
AugmentBipedalWalkerHardcore
AugmentBipedalWalkerHardcoreSmallLegs

AugmentAnt

Run main.py to create a bipedalwalker or ant environment with random morph and random control policy.

python main.py --env ENVNAME