𝒩₀-VTLA — NeoSim dual-arm
A 𝒩₀-VTLA dual-arm policy for the NeoSim simulator,
post-trained from the 𝒩₀-VTLA pretrained base.
This card documents the dual_bowl_place_stack setup.
| Embodiment | dual-arm |
| Action space | joint, 16-dim (2 x (7 joints + gripper)) |
| Action horizon | 16 |
| Tactile views | 4 |
| Latent tactile tokens | 5 |
Files
model.safetensors the full checkpoint (~8.25 GB)
assets/n0_dual_bowl_place_stack_norm/norm_stats.json
config.json architecture summary
Serve
The evaluation adapter speaks ZMQ + msgpack, so use scripts/serve_zmq.py from the
code repository:
hf download NeoteAI/n0_VTLA_dual_bowl_place_stack --local-dir checkpoints/n0_VTLA_dual_bowl_place_stack
VTLA_ASSET_ID=n0_dual_bowl_place_stack_norm python scripts/serve_zmq.py \
--config sim_dual_arm_tactile \
--ckpt checkpoints/n0_VTLA_dual_bowl_place_stack \
--addr "tcp://*:5557" \
--default-prompt "Use both arms to place and stack the bowls"
Serving reads only model.safetensors and assets/<asset-id>/norm_stats.json; no dataset is
needed. A correct load prints tactile=True with 4 views and no missing or unexpected
state-dict keys. If it reports either, the config does not match the checkpoint.
Then run the evaluation from a NeoSim checkout with a deploy YAML pointing at port 5557. Set
exec_horizon to 16, the model's action horizon: executing fewer steps clips the tail of each
chunk, which is where the grasp-closing motion lives.
Action space
Unlike the pretrained base, which predicts end-effector deltas in a canonical 32-dim rot6d container, this policy predicts joint actions: 7 joints plus gripper per arm. The joint dims are element-wise deltas against the current state; the gripper columns are absolute. Do not feed it end-effector data or reuse an end-effector normalization asset.
License
These weights are derived from Google's PaliGemma/Gemma parameters and are made available under the Gemma Terms of Use and the Gemma Prohibited Use Policy, not under the CC BY-SA 4.0 licence that covers the source code. This is inherited from the base model.