Research map / Vision-language and multimodal / Robot action and navigation
WorldAgen: Unified State-Action Prediction with Test-Time World Model Training
Chi Wan, Kangrui Wang, Yuan Si, Pingyue Zhang, Manling Li
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 test-time training of the world model with LoRA | WorldAgen · CALVIN | average sequence length: 3.87 -> 3.93 | Improved |
| add test-time training of the world model with LoRA | WorldAgen · LIBERO-10 | average success rate (%): 75.5 -> 79.0 | Improved |
| add world modeling (image prediction) to the policy (ablation) | WorldAgen · CALVIN | average sequence length: 2.96 -> 3.87 | Improved |
| add world modeling (image prediction) to the policy (ablation) | WorldAgen · LIBERO | average success rate (%): 46.5% -> 75.5% | Improved |
| vary LoRA rank from 16 to 256 during TTT (ablation) | WorldAgen-TTT · CALVIN | average sequence length: 3.928, 3.918, 3.918, 3.930, 3.923 | No clear effect |
| increase TTT data from 6 to 204 samples (ablation) | WorldAgen-TTT · CALVIN | average sequence length: 3.871 -> 3.928 | Improved |
| increase TTT data further from 204 to 340 (ablation) | WorldAgen-TTT · CALVIN | average sequence length: 3.928 -> 3.917 | Hurt |
| add Gaussian noise (std 0.1) to observations (ablation) | WorldAgen-TTT · CALVIN | average sequence length: 3.93 -> 3.90 | Hurt |
| add Gaussian noise (std 0.1) to observations (ablation) | WorldAgen-TTT · LIBERO | average success rate (%): 79.0 -> 78.0 | Hurt |
| lengthen action chunk from 5 to 9 (ablation) | WorldAgen · CALVIN | average sequence length: 3.87 -> 3.16 | Hurt |
| lengthen action chunk from 3 to 7 (ablation) | WorldAgen · LIBERO | average success rate (%): 78.0% -> 47.0% | Hurt |
| increase image chunk length from 1 to 5 (ablation) | WorldAgen · CALVIN | average sequence length: 3.87 -> 1.78 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Seer — SEER: Backdoor Detection for Vision-Language Models through Searching Target Text and Image Trigger Jointly (AAAI 2024) · better on 5 of 5 shared results
- CLOVER — Clover: Towards a Unified Video-Language Alignment and Fusion Model (CVPR 2023) · better on 1 of 1 shared result
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
- AdaWorld: Learning Adaptable World Models with Latent Actions (ICML 2025 · similarity 0.90)
- World Action Verifier: Self-Improving World Models via Forward-Inverse Asymmetry (NeurIPS 2026 · similarity 0.90)
- World–Value–Action Model: Implicit Planning for Vision–Language–Action Systems (NeurIPS 2026 · similarity 0.89)
- Reinforcing VLAs in Task-Agnostic World Models (NeurIPS 2026 · similarity 0.89)
- From Forecasting to Planning: Policy World Model for Collaborative State-Action Prediction (NeurIPS 2025 · similarity 0.88)
- WorldVLN: Autoregressive World Action Model for Aerial Vision-Language Navigation (NeurIPS 2026 · similarity 0.88)
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.
