alexhegit/so101-simstudio-lab01-pnp-molmoact2

SO-101 Lab01 Pick-and-Place — MolmoAct2

MolmoAct2 policy fine-tuned from lerobot/MolmoAct2-SO100_101-LeRobot on expert demonstrations collected and validated with SO-101 SimStudio (MuJoCo sim2sim, leader-arm teleop).

Training data: alexhegit/so101-simstudio-lab01-pnp

This Hub revision is the MI300X run: batch 32, 10 000 steps (~320K sample updates, ~15 epochs), VLM LoRA + trainable action expert, final train loss 0.009, ~30 GB of 192 GB HBM. Wall time ~13 h 40 m on AMD Instinct MI300X (DORobot). Checkpoint 010000 / last.

Cameras used for this fine-tune (2-view warm start matching the SO100_101-LeRobot base): camera_top → cam0, camera_wrist → cam1. Lab01 joints are radians; joint frame transform is identity (no SO100 degree offsets).

Closed-loop MuJoCo success rates are not published for this checkpoint yet.

Documentation

Resource	Link
SimStudio repo	rocPAI-Forge/so101-simstudio
Lab 01 walkthrough	labs/lab01_pnp/lab01_pnp.md

Quick load

from lerobot.policies.molmoact2.modeling_molmoact2 import MolmoAct2Policy

policy = MolmoAct2Policy.from_pretrained("alexhegit/so101-simstudio-lab01-pnp-molmoact2")