Research map / Video / Action recognition
SwiftWA: An Efficient Action-Centered World-Action Model
Chaojun Ni, Xinyu Zhou, YuKun Zhou, Jingyu Liu, Xiaofeng Wang, Zheng Zhu, Yang Wang, Qiuping Deng, Yun Ye, Hao Li, Zhichao Liu, Jindi Lv and 5 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add future-video prediction supervision (stride 12) during training (ablation) | GigaWorld-Policy · Real-world tasks | success rate: 0.60 -> 0.83 | Improved |
| predict too many (stride 4) or too few (stride 48) future frames instead of stride 12 (ablation) | GigaWorld-Policy · Real-world tasks | success rate: 0.83 -> 0.76 | Hurt |
| add embodied data pre-training on top of video-model initialization (ablation) | GigaWorld-Policy · Real-world tasks | success rate: 0.57 -> 0.83 | Improved |
| initialize from pretrained video model instead of scratch (ablation) | GigaWorld-Policy · Real-world tasks | success rate: 0.45 -> 0.57 | Improved |
| increase embodied pre-training data fraction (ablation) | GigaWorld-Policy · Real-world tasks | success rate: 57% -> 83% | Improved |
| use causal attention mask instead of full self-attention (ablation) | GigaWorld-Policy · Real-world tasks | success rate: 0.81 -> 0.83 | No clear effect |
| use causal attention mask instead of full self-attention (ablation) | GigaWorld-Policy · Real-world held-out video | PSNR: 27.87 -> 28.41 | Improved |
| decode actions only at inference (skip video rollout) vs Motus | GigaWorld-Policy · Inference latency | latency (ms): 3231 -> 360 | Improved |
| use GigaWorld-Policy instead of Motus | GigaWorld-Policy · RoboTwin 2.0 average | success rate: 0.89 -> 0.87 (clean) | Hurt |
| use GigaWorld-Policy instead of Motus | GigaWorld-Policy · Real-world tasks | success rate: 0.76 -> 0.83 | Improved |
| use GigaWorld-Policy instead of pi0.5 | GigaWorld-Policy · RoboTwin 2.0 average (clean) | success rate: 0.43 -> 0.87 | Improved |
| use GigaWorld-Policy instead of Motus | GigaWorld-Policy · Real-world OOD | success rate: 0.60 -> 0.72 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Cosmos-Policy — Cosmos Policy: Fine-Tuning Video Models for Visuomotor Control and Planning (ICLR 2026) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Action Images: End-to-End Policy Learning via Multiview Video Generation (NeurIPS 2026 · similarity 0.92)
- EA-WM: Event-Aware Generative World Model with Structured Kinematic-to-Visual Action Fields (NeurIPS 2026 · similarity 0.92)
- World Action Verifier: Self-Improving World Models via Forward-Inverse Asymmetry (NeurIPS 2026 · similarity 0.91)
- Beyond World-Frame Action Heads: Motion-Centric Action Frames for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.90)
- SyncWorld: Visual Calibration Enables World Models as Zero-Shot Simulators (NeurIPS 2026 · similarity 0.90)
- DiT4DiT: Jointly Modeling Video Dynamics and Actions for Generalizable Robot Control (NeurIPS 2026 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
