Fine-tunes and evaluates OpenVLA-OFT and OpenVLA-OFT+ policies for robot action generation with continuous action heads, LoRA adaptation, and FiLM conditioning on LIBERO simulation and ALOHA real-world setups.
SkillScore91Maintained
Automatic
◆AI Engineer pack
Last verified 8/12/2026
What it changes
Passed 18 of 20 test cases with the skill, and 5 of 20 without it.
Without skill27%
With skill91%
Checks it passed that the model alone missed
Sets ACTION_DIM to 14.
Sets NUM_ACTIONS_CHUNK to 25.
Pins PyTorch to version 2.2.0.
and 27 more in the full report
14 fixed6 passed either way2 still failing0 broke
Pass-rate lift
+64 pp
Tokens
+137%
Agent turns
0%
Response time
-45%
Model gemini-3.6-flashMeasured 8/12/2026v120 of 22 cases passed with skill · 20 comparable
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 13 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 13
Prompt
I have started the deploy server on my GPU workstation. What command should I execute on the client robot laptop to conduct 50 evaluation rollouts using server inference with 25 open-loop steps?
What was checked, without and with the skill
Checked
Without skill
With skill
Runs experiments/robot/aloha/run_aloha_eval.py on the client.
Missed
Passed
Passes --use_vla_server True and --num_open_loop_steps 25.
Missed
Passed
Passes --num_rollouts_planned 50.
Missed
Passed
case-12 · gemini-3.6-flash · 8/12/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md431 lines5 files
OpenVLA-OFT
Fine-tuning and evaluation workflows for OpenVLA-OFT and OpenVLA-OFT+ from the official openvla-oft codebase. Covers blank-machine setup plus LoRA-based adaptation of OpenVLA for robot action generation with continuous action prediction heads.
Quick start
Clone the public repo, follow the official setup, then evaluate a pretrained LIBERO checkpoint:
What OpenVLA-OFT changes: Standard OpenVLA tokenizes continuous actions into discrete bins, losing precision. OFT replaces this with dedicated continuous action heads (L1 regression or diffusion) while keeping the VLA backbone frozen and adapting via LoRA.
OFT vs OFT+ variants:
Variant
FiLM
Images
Typical use
OFT
Off
2 (front + wrist)
LIBERO simulation
OFT+
On
3 (high + left + right wrist)
ALOHA real-world
Key architecture choices:
LoRA adaptation: Rank-32 LoRA on VLA backbone (no full fine-tuning needed)
Continuous actions: L1 regression head (default) or diffusion head
FiLM conditioning: Feature-wise Linear Modulation for stronger language grounding in OFT+
Multi-image input: Configurable 2 or 3 camera streams via num_images_in_input