Hyfydy baselines

We include several pretrained baselines for Hyfydy. They are similar to the ones trained for our preprint. The baselines includes straight walking for sconewalk_h0918-v1, running for sconerun_h0918-v1 and similar for the other models. There is also an example for OpenSim sconewalk_h0918_osim-v1

To try the baselines, you need to first install sconegym and scone. See here for installation help.

You can play with the pre-trained baselines by using the code in this section. To train agents yourself, go to the Configuration files section.

Pre-trained baselines.

environment id	description
sconewalk_h0918_osim-v1	Energy-efficient walking with the H0918 model in OpenSim (slow performance).
sconewalk_h0918-v1	Energy-efficient walking with the H0918 model.
sconewalk_h1622-v1	Energy-efficient walking with the H1622 model.
sconewalk_h2190-v1	Energy-efficient walking with the H2190 model.
sconerun_h0918-v1	Running with the H0918 model.
sconerun_h1622-v1	Running with the H1622 model.
sconerun_h2190-v1	Running with the H2190 model.

Usage example

import gym
import sconegym
import deprl

env = gym.make('sconewalk_h0918-v1')
policy = deprl.load_baseline(env)

for ep in range(5):
    obs = env.reset()
    for i in range(1000):
        action = policy(obs)
        next_obs, reward, done, info = env.step(action)
        obs = next_obs
        if done:
            break

For the other baselines, just use: env = gym.make(‘sconewalk_h2190-v1’) or env = gym.make(‘sconerun_h2190-v1’)

You can also use noisy policy steps with:

import gym
import sconegym
import deprl

env = gym.make('sconewalk_h0918-v1')
policy = deprl.load_baseline(env)

for ep in range(5):
    obs = env.reset()
    for i in range(1000):
        # we use a noisy policy here
        action = policy.noisy_test_step(obs)
        next_obs, reward, done, info = env.step(action)
        env.sim.renderer.render_to_window()
        obs = next_obs
        if done:
            break

This can affect your performance positively or negatively, depending on the task!