DUET-DINO: Simultaneous Cross-View World Modeling for Latent Planning in Robot Manipulation
Authors: Nisarga Nilavadi, Ralf Römer, Moritz Reuss, Michael Krawez, Tobias Jülg, Angela P. Schoellig, Rudolf Lioutikov, Wolfram Burgard
arXiv: 2609.10506
Project page: https://utn-air.github.io/DUET-DINO
Abstract
Action-conditioned latent world models predict future visual representations, enabling zero-shot goal-conditioned robot planning and control. However, their predictions for fine-grained spatial and rotational actions are unreliable for full 7-DoF end-effector control. To address this gap, we introduce DUET-DINO, a simultaneous cross-view latent world model that jointly learns action-conditioned predictions from static side- and wrist-camera observations through cross-view conditioning. By exploiting complementary global scene and gripper-centric information, DUET-DINO enables latent planning over the full 7-DoF action space. Across spatially diverse reach, orientation-intensive angled-reach, and multi-goal grasp-and-lift tasks, DUET-DINO consistently outperforms single-view and independent dual-view baselines, achieving 92% success on reach, 72.5% on angled-reach, and 60.0% on lift tasks. DUET-DINO is trained from scratch on DROID and RoboArena datasets and generalizes robustly under visual distribution shifts. We further show that while V-JEPA 2 wrist-view predictions underestimate visual dynamics induced by fine-grained actions, DINOv3 predictions better capture action-conditioned scene changes, leading to stronger downstream planning. The code and model checkpoints will be open-sourced. Project page: https://utn-air.github.io/DUET-DINO
Key Results
- 92% success rate on spatially diverse reach tasks
- 72.5% success rate on orientation-intensive angled-reach tasks
- 60.0% success rate on multi-goal grasp-and-lift tasks
- Consistently outperforms single-view and independent dual-view baselines
- Robust generalization under visual distribution shifts
- Cross-view conditioning: jointly models static side-camera and wrist-camera observations
- Full 7-DoF latent planning: combines global scene context with gripper-centric information
- Backbone comparison: DINOv3 predictions capture action-conditioned scene changes better than V-JEPA 2, whose wrist-view predictions underestimate fine-grained visual dynamics
- Training data: trained from scratch on DROID and RoboArena datasets