tangger/lerobot_piper
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

backup wip

Browse Source
This commit is contained in:
Alexander Soare
2024-04-02 19:13:49 +01:00
parent 2b928eedd4
commit 65ef8c30d0
3 changed files with 80 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
lerobot/common/policies/act/policy.py
View File

@@ -54,7 +54,7 @@ class ActionChunkingTransformerPolicy(AbstractPolicy):
Args:
vae: Whether to use the variational objective. TODO(now): Give more details.
temporal_agg: Whether to do temporal aggregation. For each timestep during rollout, the action
returned as an exponential moving average of previously generated actions for that timestep.
returned as an exponential moving average of previously generated actions for that timestep.
n_obs_steps: Number of time steps worth of observation to use as input.
horizon: The number of actions to generate in one forward pass.
kl_weight: Weight for KL divergence. Defaults to None. Only applicable when using the variational
@@ -120,7 +120,7 @@ class ActionChunkingTransformerPolicy(AbstractPolicy):
"action": action.to(self.device, non_blocking=True),
}
return out
start_time = time.time()
batch = replay_buffer.sample(batch_size)