Bind-WAM: Binding-Uncertainty-Guided Selective Future Imagination in World Action Models
Organizations: Fudan University · ROBOTOPIA · Tongji University · Shanghai Innovation Institute
Abstract
World Action Models (WAMs) can improve robot learning by jointly predicting future visual states and actions, but generating a visual future at every replanning step is expensive and sometimes unnecessary. This raises a critical question: when should a robot policy imagine the future? In this paper, we propose Bind-WAM, a binding-uncertainty-guided WAM for selective future imagination. Bind-WAM introduces language-object binding uncertainty, selectively invoking joint future-action generation when the current observation does not support stable task-relevant action generation. An offline labeling pipeline uses complete demonstrations and action context to identify task-relevant objects and detect reference mismatch, invisibility, and identity drift. The resulting pseudo-labels supervise a lightweight router that predicts binding uncertainty from the current observation and language instruction. At inference, a shared WAM policy invokes selective future imagination according to the predicted uncertainty and its change between consecutive replanning steps. Simulation and real-world experiments demonstrate that Bind-WAM can establish binding uncertainty as a strong, interpretable signal for selective future imagination.
Figures & tables
| Metric | LIBERO | RoboTwin |
|---|---|---|
| Balanced Accuracy | 87.59% | 96.75% |
| AUROC | 0.9653 | 0.9984 |
| AUPRC | 0.9917 | 0.9999 |
| Brier score | 0.0504 | 0.0070 |
| Uncertain recall @ 0.5 | 97.32% | 99.27% |
| Certain recall @ 0.5 | 77.86% | 94.23% |
| Task | GO-1 | X-VLA | LingBot-VA | Motus | AdaWAM | Fast-WAM † | Fast-WAM-Joint † | Bind-WAM | ||
|---|---|---|---|---|---|---|---|---|---|---|
| Clean | ||||||||||
| Hanging Mug | 0.0 | 14.0 | 18.0 | 23.0 | 40.0 | 38.0 | 59.0 | 57.0 | 72.0 | 71.0 |
| Pick Diverse Bottles | 61.0 | 69.0 | 81.0 | 58.0 | 89.0 | 90.0 | 87.0 | 82.0 | 86.0 | 87.0 |
| Put Object Cabinet | 60.0 | 85.0 | 80.0 | 46.0 | 80.0 | 88.0 | 96.0 | 94.0 | 95.0 | 95.0 |
| Rotate QR Code | 22.0 | 74.0 | 89.0 | 34.0 | 96.0 | 89.0 | 94.0 | 93.0 | 91.0 | 92.0 |
| Scan Object | 1.0 | 55.0 | 72.0 | 14.0 | 96.0 | 67.0 | 92.0 | 89.0 | 92.0 | 89.0 |
World Action Models
World Action Models connect predictions of how scenes change under action to executable policies for robots, vehicles, and other embodied agents.
See all 241 papers