openwam

OpenWAM world-action model trained on RoboDojo (predict_future): predicted camera views and end-effector motion

• Kind: world_model
• Protocol: http
• Default port: 8820
• VRAM: 15.3 GB
• Load time: about 240 s
• Skills that use a world_model: wam
• Configurations that name it: –

What the model is told

OpenWAM world-action model trained on RoboDojo: predicts the camera views and end-effector motion for an instruction

Weights

harness fetch openwam --deploy <machine>

Directory	Source	From
<weights>/openwam	hf_model	OpenWAM/OpenWAM-Alpha-Sim-RoboDojo

Notes

Uses XPolicyLab’s OpenWAM adapter (policy/OpenWAM/model.py) to convert the predicted actions.

Manifest

kind = "world_model"
summary = "OpenWAM world-action model trained on RoboDojo (predict_future): predicted camera views and end-effector motion"
protocol = "http"
command = ["{python}", "-u", "services/servers/serve_openwam.py", "--ckpt-dir", "{weights}/openwam", "--port", "{port}"]
port = 8820
# Measured on a 3090 with the UMT5-XXL text encoder on the CPU: 15.3 GB.
vram_gb = 15.3
load_seconds = 240
env = { HARNESS_ROBODOJO_ROOT = "{robodojo}" }
client = { url = "{url}", description = "OpenWAM world-action model trained on RoboDojo: predicts the camera views and end-effector motion for an instruction" }
needs_files = ["{weights}/openwam/config.yaml", "{xpolicylab}/policy/OpenWAM/model.py"]
notes = "Uses XPolicyLab's OpenWAM adapter (policy/OpenWAM/model.py) to convert the predicted actions."

[[weights]]
dir = "openwam"
hf_model = "OpenWAM/OpenWAM-Alpha-Sim-RoboDojo"