jaco

Exploring the Jaco 6DOF arm state returns in Mujoco

state_dim = env.observation_spec()['observations'].shape[0] action_dim = env.action_spec().shape[0] min_action = float(env.action_spec().minimum[0]) max_action = float(env.action_spec().maximum[0]) action_shape = env.action_spec().shape

state return in mujoco -

---

after reset:

state_type, reward, discount, state = eval_env.reset()

state_type is : <StepType.FIRST: 0>

reward, discount are None

state is: collections.OrderedDict['observations'].shape -> (37,) collections.OrderedDict['observations'].dtype -> np.float64 max() is 1, min() -1 -1

action was of shape (6,) float32

---

during steps

step_type, reward, discount, state = eval_env.step(action) step_type is: <StepType.MID: 1> done is bool discount was 1.0 reward was 0.0 - (obv this will change) observations has same params as above