admin/lerobot
1
0
Fork 0
mirror of https://github.com/huggingface/lerobot.git synced 2026-05-17 17:50:09 +00:00
Code Issues Packages Projects Releases Wiki Activity

chore(training): widen augmentation envelope after live-robot diagnostic

Browse Source 
The tensor-level comparison between dry-run (dataset frame) and live-
robot inference proved the runtime is bug-free — same shape, dtype,
device, channel order, batch dim, and normalization on both paths.
The remaining variable: front-camera mean brightness was 0.26 live vs
0.39 on the dataset frame, ~33% darker. Training augmentation only
covered ±20% brightness, so the live scene sits just outside the
supervised envelope and the LM head collapses to its dominant prior.

Widen the augmentation knobs for the next retrain:

  * brightness    0.8–1.2  → 0.5–1.6   (covers ~30% darker / 60% lighter)
  * contrast      0.8–1.2  → 0.6–1.5
  * saturation    0.5–1.5  → 0.3–1.7
  * hue          ±0.05    → ±0.10
  * affine        ±5°/±5%  → ±15°/±15% (covers cube placement / camera drift)
  * max_num_transforms 3 → 4

And bump prompt-component dropout (subtask 0.20 → 0.30) so the LM
can't lean on stale memorised plan/memory at inference.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
Pepijn
2026-05-12 18:25:41 +02:00
parent 53172873e3
commit 47fb8318b1
1 changed files with 10 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
examples/training/smolvla2_hirobot.slurm
+10 -5
View File
@@ -52,7 +52,7 @@ echo " GPUs: $NUM_PROCESSES"
echo " batch: $BATCH_SIZE / GPU (global=$((NUM_PROCESSES * BATCH_SIZE)))"
echo " steps: $STEPS"
echo " output: $OUTPUT_DIR"
echo " augmentation: image_transforms ON, prompt dropout {plan:0.15 memory:0.15 subtask:0.20}"
echo " augmentation: image_transforms ON (wide), prompt dropout {plan:0.20 memory:0.20 subtask:0.30}"
accelerate launch --multi_gpu --num_processes="$NUM_PROCESSES" \
-m lerobot.scripts.lerobot_train \
@@ -62,11 +62,16 @@ accelerate launch --multi_gpu --num_processes="$NUM_PROCESSES" \
--dataset.revision=main \
--dataset.video_backend=pyav \
--dataset.image_transforms.enable=true \
--dataset.image_transforms.max_num_transforms=3 \
--dataset.image_transforms.max_num_transforms=4 \
--dataset.image_transforms.random_order=true \
--policy.plan_dropout_prob=0.15 \
--policy.memory_dropout_prob=0.15 \
--policy.subtask_dropout_prob=0.20 \
--dataset.image_transforms.tfs.brightness.kwargs='{"brightness": [0.5, 1.6]}' \
--dataset.image_transforms.tfs.contrast.kwargs='{"contrast": [0.6, 1.5]}' \
--dataset.image_transforms.tfs.saturation.kwargs='{"saturation": [0.3, 1.7]}' \
--dataset.image_transforms.tfs.hue.kwargs='{"hue": [-0.1, 0.1]}' \
--dataset.image_transforms.tfs.affine.kwargs='{"degrees": [-15.0, 15.0], "translate": [0.15, 0.15]}' \
--policy.plan_dropout_prob=0.20 \
--policy.memory_dropout_prob=0.20 \
--policy.subtask_dropout_prob=0.30 \
--output_dir="$OUTPUT_DIR" \
--job_name="$JOB_NAME" \
--policy.repo_id="$POLICY_REPO_ID" \