hangX/lerobot
1
0
Fork 0
forked from tangger/lerobot
Code Pull Requests Activity

[HIL-SERL] Update CI to allow installation of prerelease versions for lerobot (#1018)

Browse Source 
Co-authored-by: imstevenpmwork <steven.palma@huggingface.co>
This commit is contained in:
Adil Zouitine
2025-04-24 10:18:03 +02:00
committed by GitHub
parent a0018240d5
commit 299effe0f1
10 changed files with 78 additions and 388 deletions
Download Patch File Download Diff File Expand all files Collapse all files

45
lerobot/common/envs/utils.py
View File

@@ -37,35 +37,29 @@ def preprocess_observation(observations: dict[str, np.ndarray]) -> dict[str, Ten
"""
# map to expected inputs for the policy
return_observations = {}
# TODO: You have to merge all tensors from agent key and extra key
# You don't keep sensor param key in the observation
# And you keep sensor data rgb
for key, img in observations.items():
if "images" not in key:
continue
if "pixels" in observations:
if isinstance(observations["pixels"], dict):
imgs = {f"observation.images.{key}": img for key, img in observations["pixels"].items()}
else:
imgs = {"observation.image": observations["pixels"]}
# TODO(aliberts, rcadene): use transforms.ToTensor()?
if not torch.is_tensor(img):
for imgkey, img in imgs.items():
# TODO(aliberts, rcadene): use transforms.ToTensor()?
img = torch.from_numpy(img)
if img.ndim == 3:
img = img.unsqueeze(0)
# sanity check that images are channel last
_, h, w, c = img.shape
assert c < h and c < w, f"expect channel last images, but instead got {img.shape=}"
# sanity check that images are channel last
_, h, w, c = img.shape
assert c < h and c < w, f"expect channel last images, but instead got {img.shape=}"
# sanity check that images are uint8
assert img.dtype == torch.uint8, f"expect torch.uint8, but instead {img.dtype=}"
# sanity check that images are uint8
assert img.dtype == torch.uint8, f"expect torch.uint8, but instead {img.dtype=}"
# convert to channel first of type float32 in range [0,1]
img = einops.rearrange(img, "b h w c -> b c h w").contiguous()
img = img.type(torch.float32)
img /= 255
# convert to channel first of type float32 in range [0,1]
img = einops.rearrange(img, "b h w c -> b c h w").contiguous()
img = img.type(torch.float32)
img /= 255
return_observations[key] = img
# obs state agent qpos and qvel
# image
return_observations[imgkey] = img
if "environment_state" in observations:
return_observations["observation.environment_state"] = torch.from_numpy(
@@ -74,8 +68,7 @@ def preprocess_observation(observations: dict[str, np.ndarray]) -> dict[str, Ten
# TODO(rcadene): enable pixels only baseline with `obs_type="pixels"` in environment by removing
# requirement for "agent_pos"
# return_observations["observation.state"] = torch.from_numpy(observations["agent_pos"]).float()
return_observations["observation.state"] = observations["observation.state"].float()
return_observations["observation.state"] = torch.from_numpy(observations["agent_pos"]).float()
return return_observations
@@ -93,7 +86,7 @@ def env_to_policy_features(env_cfg: EnvConfig) -> dict[str, PolicyFeature]:
else:
feature = ft
policy_key = env_cfg.features_map.get(key, key)
policy_key = env_cfg.features_map[key]
policy_features[policy_key] = feature
return policy_features