Add independent DreamWaQ stair capability benchmark
This commit is contained in:
@@ -47,10 +47,10 @@ MAX_VX, MAX_VY, MAX_WZ = 1.0, 1.0, 1.0
|
||||
# MuJoCo actuator and qpos order: FR, FL, RR, RL; hip, thigh, calf.
|
||||
DEFAULT_ANGLES = np.array(
|
||||
[
|
||||
0.0, 0.9, -1.8,
|
||||
0.0, 0.9, -1.8,
|
||||
0.0, 0.9, -1.8,
|
||||
0.0, 0.9, -1.8,
|
||||
-0.1, 0.8, -1.5,
|
||||
0.1, 0.8, -1.5,
|
||||
-0.1, 1.0, -1.5,
|
||||
0.1, 1.0, -1.5,
|
||||
],
|
||||
dtype=np.float32,
|
||||
)
|
||||
@@ -272,7 +272,7 @@ def main():
|
||||
row = int(np.clip(((y - geom_position[1]) / size_y * 0.5 + 0.5) * (rows - 1), 0, rows - 1))
|
||||
return float(geom_position[2] + base + samples[row, column] * height)
|
||||
|
||||
spawn_z = hfield_z(spawn_x, spawn_y) + 0.45
|
||||
spawn_z = hfield_z(spawn_x, spawn_y) + 0.34
|
||||
|
||||
def reset_state():
|
||||
mujoco.mj_resetData(model, data)
|
||||
@@ -318,6 +318,7 @@ def main():
|
||||
last_action = np.zeros(NUM_ACTIONS, dtype=np.float32)
|
||||
action = np.zeros(NUM_ACTIONS, dtype=np.float32)
|
||||
history = np.zeros((1, HISTORY_LEN, NUM_OBS), dtype=np.float32)
|
||||
history_initialized = False
|
||||
decimation = 4
|
||||
physics_step = 0
|
||||
reached_top = False
|
||||
@@ -335,6 +336,7 @@ def main():
|
||||
reset_state()
|
||||
last_action.fill(0.0)
|
||||
history.fill(0.0)
|
||||
history_initialized = False
|
||||
reached_top = False
|
||||
fallen_since = None
|
||||
next_log_time = 0.0
|
||||
@@ -352,8 +354,9 @@ def main():
|
||||
if physics_step % decimation == 0:
|
||||
command = np.array([vx, vy, wz], dtype=np.float32)
|
||||
obs = compute_obs(model, data, command, last_action)
|
||||
history[:, :-1] = history[:, 1:]
|
||||
history[:, -1] = obs
|
||||
if not history_initialized:
|
||||
history[:, -1] = obs
|
||||
history_initialized = True
|
||||
action = session.run(
|
||||
None,
|
||||
{
|
||||
@@ -367,6 +370,9 @@ def main():
|
||||
break
|
||||
action = np.clip(action, -CLIP_ACTIONS, CLIP_ACTIONS)
|
||||
last_action = action.copy()
|
||||
if physics_step > 0:
|
||||
history[:, :-1] = history[:, 1:]
|
||||
history[:, -1] = obs
|
||||
|
||||
target = DEFAULT_ANGLES + action * ACTION_SCALE
|
||||
actuator_joints = model.actuator_trnid[:, 0]
|
||||
|
||||
Reference in New Issue
Block a user