Piper Pick-Cup Left-Arm Diffusion Policy
This repository contains the compact EMA inference artifact for a
single-episode, real-world Piper pick-cup Diffusion Policy run.
Artifact
- File:
piper_pick_cup_left_ema.pt
- Format:
adavip.dp_sanity.inference.v1
- SHA-256:
da7d8b63c0f28df3e8534e9bc47d848af7a207be31af0032a593e05fde2512a2
- Size: 1,316,987,288 bytes
- Weights: EMA policy with its normalizer and resolved policy config
Contract
- Control frequency: 30 Hz
- Observation history: 2 frames
- Observations:
cam_high [3,480,640], cam_left_wrist [3,480,640],
qpos [7]
- Output: 15 left-arm actions, each with 6 joints and 1 gripper dimension
- Image format: RGB float32 in
[0,1]
The artifact contains only the learned inference policy. Robot communication,
safety checks, observation synchronization, and action execution must be
provided by the deployment application.
