Sim2Real prototype example

[jgillick]

A simple example of how the environment might be used in a robot deployment.

• Within the environment, rewards and terminations are disabled based on the mode value.
• Within eval_deploy.py
  • Observation config functions are replaced with functions that could return actual values (currently just zeros)
  • The latest action actuator values are fetched and could be passed to the actuators.

Override observation functions:

    obs = env.observation_manager.cfg
    obs["angle_velocity"].fn = lambda env: torch.zeros(3)
    obs["linear_velocity"].fn = lambda env: torch.zeros(3)
    obs["projected_gravity"].fn = lambda env: torch.zeros(3)
    obs["dof_position"].fn = lambda env: torch.zeros(12)
    obs["dof_velocity"].fn = lambda env: torch.zeros(12)

Get actuator values after performing step:

obs, _rews, _dones, _infos = env.step(actions)
actions = env.action_manager.get_actions()