run_the_baselines.md

Run the baselines

This year we provide several baselines:

• Manipulation track and locomotion track baselines with deprl, also see the docs.
• A reflex-based locomotion controller, see here.

These baselines will not give you good task performance or win the challenge for you, but they provide a nice starting point.

To run the deprl-baselines, you need to install:

pip install deprl

Take a look here if you run into issues or want to install torch-cpu. The requirements for the reflex-based baseline are contained in the above link.

Manipulation Track

This deprl-baseline will try to lift the cube upwards.

import gym
import myosuite, deprl

env = gym.make('myoChallengeRelocateP1-v0')
policy = deprl.load_baseline(env)

for ep in range(5):
    print(f'Episode: {ep} of 5')
    state = env.reset()
    while True:
        action = policy(state)
        # uncomment if you want to render the task
        # env.mj_render()
        next_state, reward, done, info = env.step(action)
        state = next_state
        if done: 
            break

You can also use policy.noisy_test_step(state) for actions with Gaussian noise. Your results may vary!

Locomotion track

This deprl-baseline will try to stand around and slowly move across the quad.

import gym
import myosuite, deprl

env = gym.make('myoChallengeChaseTagP1-v0')
policy = deprl.load_baseline(env)

for ep in range(5):
    print(f'Episode: {ep} of 5')
    state = env.reset()
    while True:
        action = policy(state)
        # uncomment if you want to render the task
        # env.mj_render()
        next_state, reward, done, info = env.step(action)
        state = next_state
        if done: 
            break