Downloads · 30 days
0
Zhiyuan17/franka-table-top-0904-pi05-16k
franka-table-top-0904-pi05-16k is a robotics model from Zhiyuan17. Use it for the robotics task on the model card, and read the license before you ship it in a product. It is set up for openpi. The card lists the license as other.
Final full-finetuning checkpoint for the one-camera Franka tabletop experiments. The model was initialized from the same FP32-converted official π0.5 base used by the associated LIBERO and RoboCasa experiments and tra…
Downloads · 30 days
0
Access
Public
Updated Sep 7, 2026
Parameters
3.6B
7.5 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors7.5 GB · 100%
How the weights are stored.
BF163.5B · 97%
From the Hugging Face model README
Final full-finetuning checkpoint for the one-camera Franka tabletop experiments. The model was initialized from the same FP32-converted official π0.5 base used by the associated LIBERO and RoboCasa experiments and trained for 16,000 optimizer updates.
model.safetensors: inference model weights.checkpoint_manifest.json: immutable architecture, schema, provenance, size,
and SHA256 metadata.assets/franka_table_top_0904_pi05_native_v1/norm_stats.json: matching state
and action quantile normalization.Optimizer, RNG, and data-loader continuation state are intentionally omitted.
Use the companion
franka-pi05-real-robot
deployment repository to download, validate, and serve this checkpoint.
[ee_xyz_m, quaternion_xyzw, gripper_width_m].[delta_xyz_m, angular_xyz_rad, gripper_binary_target] at 10 Hz.discrete_state_input=false.The exact live integration convention for the three angular action values must be verified against the robot controller before physical rollout. This model must be used with workspace bounds, per-step limits, an emergency stop, and a human operator. Low training loss does not establish safe or successful robot behavior.