Model: Motus — unified latent-action world model (WAM+VLA hybrid) · Tsinghua University (THU-ML) & collaborators (Jun Zhu / Hang Su group; incl. Horizon Robotics) Paper: arXiv 2512.13030 (v1 Dec 15 2025 · v2 Dec 25 2025) · open weights + code (GitHub thu-ml/Motus, HF motus-robotics/Motus, project page) Status: arXiv preprint (not yet peer-reviewed) — filed under Latest Papers. Authors: Hongzhe Bi, Hengkai Tan, Shenghao Xie, Zeyuan Wang, … Zhizhong Su, Lei Ma, Hang Su, Jun Zhu.

One of the concrete instances of the WAM+VLA hybrid NVIDIA's blog predicts (NVIDIA WAM thesis §2.6) — a single Mixture-of-Transformers that can be a world model, a VLA, an inverse-dynamics model, or a video generator on demand. Companion: VLA Architectures §4.2b · World Models · DYNA-2 · Being-H0.7.

1. One model, five modes. Motus is a Mixture-of-Transformers (MoT) integrating three experts — understanding, video generation, action — with a UniDiffuser-style scheduler that flexibly switches among world model · VLA · inverse-dynamics model · video generation · video-action joint prediction. Unifying all of these in one pretraining is the contribution.
2. Latent actions from optical flow. It learns latent actions by extracting a pixel-level "delta action" from optical flow, which lets it pretrain action representations at scale from action-free video (human + multi-robot).
3. 8B params across four towers: video-generation model 5.00B, VLM 2.13B, action expert 641.5M, understanding expert 253.5M.
4. SOTA on RoboTwin 2.0. On the RoboTwin 2.0 randomized multi-task setting: 88.66% vs X-VLA 72.80% and π0.5 42.98% (= +15% over X-VLA, +45% over π0.5). Real-world (two platforms, 9 tasks): +11–48% over π0.5.

• It operationalizes the "unified" corner of the WAM+VLA hybrid. Where DYNA-2 drops its video tower at inference (reactive) and π0.7 bolts a world model onto a VLA modularly, Motus keeps all modes live in one MoT and schedules which to run — the most literal reading of "collapse WAM and VLA into one pretraining."
• Optical-flow latent actions are a cheap, general action-supervision bridge. Like DYNA-2's hand-pose pseudo-actions, this turns action-free video into trainable action signal — but via a task-agnostic motion cue (flow) rather than hand tracking, so it extends to non-hand, multi-robot footage.
• Open weights + code. Unlike DYNA-2 (closed) or Cosmos 3 (partial), Motus is fully released, so its unified-MoT recipe is independently checkable.

Mixture-of-Transformers, four towers, one scheduler. Each modality is a specialized expert; the UniDiffuser-style scheduler sets per-modality noise levels so the same weights realize different models:

flowchart LR
  IN[obs · language] --> UND[Understanding expert · 253M]
  IN --> VGM[Video-generation expert · 5.0B]
  IN --> ACT[Action expert · 641M]
  VLM[VLM · 2.13B] --- UND
  UND <-. shared self-attention .-> VGM
  VGM <-. shared self-attention .-> ACT
  UND <-. shared self-attention .-> ACT
  SCHED[UniDiffuser scheduler<br/>sets per-modality noise level] -. selects mode .-> VGM
  SCHED -. selects mode .-> ACT
  ACT ==> OUT[action]

Latent actions via optical flow. Motus derives a pixel-level "delta action" from optical flow between frames, giving an embodiment-agnostic latent action label learnable from any video — the substrate for large-scale action pretraining.

Three-phase training over a six-layer data pyramid.

• Phase 1 — Learning Visual Dynamics: adapt the video-generation model on multi-robot and human videos.
• Phase 2 — Learning Action Representations: pretrain the unified model with the optical-flow latent actions.
• Phase 3 — Specializing for the Target Robot: fine-tune on target-robot data.
• Data pyramid (quantity ↓, quality ↑): web data → egocentric human video → synthetic → task-agnostic → multi-robot trajectories → target-robot task data.

• Real-world, two robot platforms, 9 tasks (fold towel · brew coffee · grind beans · pour water · touch keyboard · grab cube · place cube · get water from dispenser · put bread in oven): +11–48% over π0.5.
• Ablations report that unifying all functionalities/priors in one model is what drives the downstream gains (vs single-mode baselines).

Significance. Motus is the clearest open demonstration that a single scheduled MoT can serve every world/action role — the unified endpoint of Category E (Review-VLA-Architecture §4.2b). The optical-flow latent-action recipe is a general alternative to hand-pose (DYNA-2) or reconstruction-free latents (ω-0).

Limitations.

1. arXiv preprint, unreplicated. Numbers are the authors' own; no peer review yet.
2. No explicit limitations/failure section. Future work only ("more universal motion priors; latent actions from internet-scale video").
3. Optical flow as action proxy is coarse — flow conflates camera and object/effector motion; how well "delta action" grounds fine contact-rich control isn't isolated.
4. RoboTwin 2.0 is simulation; real-world evidence is a 9-task, two-platform demo, not a broad hardware study.

• Paper: arXiv 2512.13030 · code/weights: GitHub · HF · project
• VLA Architectures §4.2b (WAM×VLA hybrids) · World Models
• Sibling hybrids: DYNA-2 · ω-0 · Being-H0.7 · Cortex 2.0 · Cosmos 3 / NVIDIA WAM
• Latest Papers

← Back to Latest Papers · Home · Reviews