OpenWAM IDM: Mixed CF64k
Project | Code | Inference guide
From visual plans to robot actions. OpenWAM IDM is an inverse dynamics model that predicts robot controls from a current observation and a desired visual future. It separates planning from control: a video predictor describes the motion, and the IDM translates that motion into actions.
This checkpoint combines 60% counterfactual dynamics and 40% real demonstrations during IDM post-training on LIBERO-Long (LIBERO-10). The CF64k collection provides varied action interventions alongside the task-directed motions in demonstrations. Built on OpenWAM's robot-video-pretrained backbone, the model is trained for 20,000 post-training steps.
The IDM uses the current agent and wrist camera views, robot proprioception, and a short future video represented as latents. It predicts end-effector delta controls and gripper commands, without task-language conditioning or past actions. The same frozen IDM can therefore serve as the action component for different task-adapted video predictors.
Evaluation
| Evaluation | Success |
|---|---|
| LIBERO-10, 500 episodes | 95.4% |
| Transfer to four LIBERO-90 tasks, 50 episodes each | 94.0% |
Results measure the composed video-to-action system with this IDM frozen. LIBERO-10 uses the OpenWAM VTA step-9700 video predictor (477/500 successes). Transfer uses task-tuned video predictors on tasks 21, 45, 64, and 74; the reported score is the mean across those four tasks.
Use with OpenWAM
Pair this checkpoint with a video predictor through OpenWAM's video-to-action pipeline. The release includes model weights and an inference configuration; the guide covers setup, shared frontend assets, and evaluation.
For the counterfactual-only counterpart, see OpenWAM IDM CF64k.
License
Model weights are released under AGPL-3.0-only. Upstream attribution and license terms are retained in NOTICE and LICENSE.upstream.