Driving · 2026-10-01
DiffWAM: A Fast and Efficient Navigation World Action Model
Mo Zhu, Yuze Wu, Xijie Huang, Xiao Cui, Fei Gao, Xin Zhou
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
DiffWAM is a geometry-conditioned navigation world-action model that decodes continuous UAV camera trajectories directly from intermediate predictive features of a frozen video foundation model. It uses Grid-Motion and Latent2Pose with first-frame geometry, and needs complete future-video rollout + geometric reconstruction only for offline supervision. FastDreamer further overlaps predictive/geometry computation with flight-time execution and uses timestamp-aware trajectory handoff.
Why it matters
The authors target a deployment bottleneck in video-foundation navigation: expensive future-video synthesis and multi-frame reconstruction. DiffWAM replaces a generate-then-reconstruct pipeline with a direct predictive-to-motion interface, grounding motion-relevant representations into metrically meaningful 3D camera trajectories using first-frame geometry, and adds an inference/execution mechanism for continuous UAV operation.
Method
- Direct predictive-to-motion: DiffWAM maps early predictive representations from a frozen video model into time-indexed camera poses instead of decoding future RGB frames.
- Geometry grounding: Grid-Motion preserves spatial-temporal motion associations, and Latent2Pose uses first-frame geometry from frozen MoGe2 to recover metrically meaningful 3D motion.
- Continuous execution: FastDreamer overlaps predictive/geometry computation with ongoing flight and performs timestamp-aware asynchronous trajectory handoff for continuous UAV execution.
Limitation
The paper states that further evaluation should quantify task-specific closed-loop completion, physical-scale accuracy, and complete navigation-update latency across unseen environments.
Abstract (from arXiv)
Pretrained video foundation models encode rich semantic and spatiotemporal priors for embodied navigation, yet converting these priors into UAV motion typically requires expensive future-video synthesis and geometric reconstruction. We investigate whether the motion implicit in future visual prediction can instead be recovered directly from the predictive representations of a frozen video model. To this end, we present DiffWAM, a geometry-conditioned navigation world-action model that directly transforms multi-level predictive features into continuous camera trajectories. Its Grid-Motion module preserves spatial-temporal motion associations, while Latent2Pose grounds them with first-frame geometry to recover metrically meaningful 3D motion. Complete video rollouts and geometric reconstruction are required only for offline supervision, eliminating future-video decoding and multi-frame reconstruction during deployment. We further introduce FastDreamer, which overlaps predictive and geometric computation with ongoing flight and performs timestamp-aware asynchronous trajectory handoff for continuous UAV execution. DiffWAM achieves a trajectory RMSE of 0.3492 m and an endpoint success rate of 74.40% on the 1,000-sample DiffWAM-1000 benchmark, while representative real-world experiments demonstrate complex behaviors including constrained traversal, orbiting, S-shaped flight, and multi-stage navigation. An onboard DiffWAM-Flash implementation further reaches 1.08 s model-pipeline latency on NVIDIA Jetson AGX Thor. These results demonstrate that predictive video representations can be efficiently grounded into continuous 3D motion, providing a direct alternative to generate-then-reconstruct navigation pipelines. Project page: https://zzmmzzm.github.io/diffwam.github.io/.
Related papers
- G$^2$WAM: World Action Models for Efficient and Geometry Grounded Image-Goal Navigation
- De-Diffusing WAMs: Action-Sufficient Direct Prediction for Robot Control
- ScaleNav: Scalable Internet Video Pretraining for Language-Conditioned Local Navigation
- Native Action-Prior Learning from Videos for World Action Models
- 1Step-WAM: Your World Action Model but Faster using 1-Step Action Generation
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-09-30 · Indexed by Pub-AI: 2026-10-01 UTC
DiffWAM: A Fast and Efficient Navigation World Action Model
Preprint posted: 2026-09-30 · Original record ↗
- arxiv
- 2609.39763
Explicit source versions (1)
- v1 · 2026-09-30
- Preprint available
Source date: 2026-09-30 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
Pretrained video foundation models encode rich semantic and spatiotemporal priors for embodied navigation, yet converting these priors into UAV motion typically requires expensive future-video synthesis and geometric reconstruction. We investigate whether the motion implicit in future visual prediction can instead be recovered directly from the predictive representations of a frozen video model. To this end, we present DiffWAM, a geometry-conditioned navigation world-action model that directly transforms multi-level predictive features into continuous camera trajectories. Its Grid-Motion module preserves spatial-temporal motion associations, while Latent2Pose grounds them with first-frame geometry to recover metrically meaningful 3D motion. Complete video rollouts and geometric reconstruction are required only for offline supervision, eliminating future-video decoding and multi-frame reconstruction during deployment. We further introduce FastDreamer, which overlaps predictive and geometric computation with ongoing flight and performs timestamp-aware asynchronous trajectory handoff for continuous UAV execution. DiffWAM achieves a trajectory RMSE of 0.3492 m and an endpoint success rate of 74.40% on the 1,000-sample DiffWAM-1000 benchmark, while representative real-world experiments demonstrate complex behaviors including constrained traversal, orbiting, S-shaped flight, and multi-stage navigation. An onboard DiffWAM-Flash implementation further reaches 1.08 s model-pipeline latency on NVIDIA Jetson AGX Thor. These results demonstrate that predictive video representations can be efficiently grounded into continuous 3D motion, providing a direct alternative to generate-then-reconstruct navigation pipelines. Project page: https://zzmmzzm.github.io/diffwam.github.io/.