Remove latency, tdmpc policy passes tests (TODO: make it work with online RL)

lerobot/configs/policy/act.yaml

@@ -10,7 +10,6 @@ log_freq: 250
 
 horizon: 100
 n_obs_steps: 1
-n_latency_steps: 0
 # when temporal_agg=False, n_action_steps=horizon
 n_action_steps: ${horizon}
 

lerobot/configs/policy/diffusion.yaml

@@ -16,7 +16,6 @@ seed: 100000
 horizon: 16
 n_obs_steps: 2
 n_action_steps: 8
-n_latency_steps: 0
 dataset_obs_steps: ${n_obs_steps}
 past_action_visible: False
 keypoint_visible_rate: 1.0
@@ -38,7 +37,6 @@ policy:
   shape_meta: ${shape_meta}
 
   horizon: ${horizon}
-  # n_action_steps: ${eval:'${n_action_steps}+${n_latency_steps}'}
   n_obs_steps: ${n_obs_steps}
   num_inference_steps: 100
   obs_as_global_cond: ${obs_as_global_cond}
@@ -64,6 +62,11 @@ policy:
   lr_warmup_steps: 500
   grad_clip_norm: 10
 
+  delta_timestamps:
+    observation.image: [-.1, 0]
+    observation.state: [-.1, 0]
+    action: [-.1, 0, .1, .2, .3, .4, .5, .6, .7, .8, .9, 1.0, 1.1, 1.2, 1.3, 1.4]
+
 noise_scheduler:
   _target_: diffusers.schedulers.scheduling_ddpm.DDPMScheduler
   num_train_timesteps: 100

lerobot/configs/policy/tdmpc.yaml

@@ -77,3 +77,9 @@ policy:
   num_q: 5
   mlp_dim: 512
   latent_dim: 50
+
+  delta_timestamps:
+    observation.image: "[i / ${fps} for i in range(6)]"
+    observation.state: "[i / ${fps} for i in range(6)]"
+    action: "[i / ${fps} for i in range(5)]"
+    next.reward: "[i / ${fps} for i in range(5)]"