pi05

pi0.5 VLA, RoboDojo’s own checkpoint, through XPolicyLab’s Pi_05 adapter (openpi, JAX)

• Kind: policy
• Protocol: xpolicylab
• Default port: 19007
• VRAM: 16 GB
• Load time: about 300 s
• Skills that use a policy: vla
• Configurations that name it: –

What the model is told

pi0.5 (RoboDojo’s own checkpoint) trained on whole RoboDojo tasks: demonstrations from the start pose, with each task’s own instruction (joint actions, both arms)

Weights

harness fetch pi05 --deploy <machine>

Directory	Source	From
<weights>/pi05	modelscope_dataset	RoboDojo-Benchmark/RoboDojo (ckpt/RoboDojo/Pi_05/RoboDojo-sim-arx_x5-joint-0)

Notes

Set OPENPI_DATA_HOME in the deploy file’s env for this service (openpi’s cache).

Manifest

kind = "policy"
summary = "pi0.5 VLA, RoboDojo's own checkpoint, through XPolicyLab's Pi_05 adapter (openpi, JAX)"
protocol = "xpolicylab"
port = 19007
vram_gb = 16.0
load_seconds = 300
# openpi's own uv environment (XPolicyLab policy/Pi_05/openpi/.venv): JAX does not share the project's.
python = "{xpolicylab}/policy/Pi_05/openpi/.venv/bin/python"
env = { XLA_PYTHON_CLIENT_PREALLOCATE = "false" }
client = { url = "{url}", description = "pi0.5 (RoboDojo's own checkpoint) trained on whole RoboDojo tasks: demonstrations from the start pose, with each task's own instruction (joint actions, both arms)" }
needs_files = ["{weights}/pi05", "{xpolicylab}/policy/Pi_05/openpi/.venv/bin/python"]
notes = "Set OPENPI_DATA_HOME in the deploy file's env for this service (openpi's cache)."

[xpolicylab]
policy = "Pi_05"
ckpt_name = "{weights}/pi05"
action_type = "joint"
pythonpath = ["{xpolicylab_parent}", "{xpolicylab}/policy/Pi_05/openpi/src"]

[[weights]]
dir = "pi05"
modelscope_dataset = "RoboDojo-Benchmark/RoboDojo"
root = "ckpt/RoboDojo/Pi_05/RoboDojo-sim-arx_x5-joint-0"