Research map / Video / Video generation
UniPhys: Unified Planner and Controller with Diffusion for Flexible Physics-Based Character Control
Yan Wu, Korrawe Karunratanakul, Zhengyi Luo, Siyu Tang
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 latent action representation (ablation) | UniPhys autoregressive denoising with stabilization · Unconditional rollout | mean episode length: 238.2 -> 2320.3 | Improved |
| add stabilization trick (ablation) | UniPhys autoregressive denoising with latent action · Unconditional rollout | mean episode length: 148.2 -> 2320.3 | Improved |
| add stabilization trick (ablation) | UniPhys autoregressive denoising with raw actions · Unconditional rollout | mean episode length: 58.0 -> 238.2 | Improved |
| use autoregressive instead of full-sequence denoising (ablation) | UniPhys · Unconditional rollout | mean episode length: 231.6 -> 2320.3 | Improved |
| use autoregressive instead of gradual denoising (ablation) | UniPhys · Unconditional rollout | mean episode length: 1817.6 -> 2320.3 | Improved |
| use gradual instead of autoregressive denoising (ablation) | UniPhys · Unconditional rollout | FPS: 9.5 -> 18.0 | Improved |
| use full-sequence instead of gradual denoising (ablation) | UniPhys · Unconditional rollout | FPS: 18.0 -> 23.0 | Improved |
| increase Monte-Carlo guidance samples from 1 to 5 (ablation) | UniPhys loss-guided sampling · Goal reaching | success rate: 26% -> 98% | Improved |
| increase Monte-Carlo guidance samples from 1 to 5 (ablation) | UniPhys loss-guided sampling · Goal reaching | FPS: 9.2 -> 8.7 | Hurt |
| use UniPhys instead of MaskedMimic | text-driven control · BABEL text control | jerk [mm/s^3]: 4.2 -> 1.2 | Improved |
| use UniPhys instead of CLoSD | text-driven control · BABEL text control | correct caption rate: 61.6% -> 56.3% | Hurt |
| use UniPhys instead of MaskedMimic | far goal reaching · Goal reaching walk-far | success rate: 0.0 -> 0.95 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- CLoSD — CLoSD: Closing the Loop between Simulation and Diffusion for multi-task character control (ICLR 2025) · better on 4 of 8 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
- CLoSD: Closing the Loop between Simulation and Diffusion for multi-task character control (ICLR 2025 · similarity 0.90)
- NaP-Control: Navigating Diffusion Prior for Versatile and Fast Character Control (ECCV 2026 · similarity 0.89)
- Pixel Motion Diffusion is What We Need for Robot Control (CVPR 2026 · similarity 0.87)
- Zero-Shot Robotic Manipulation with Pre-Trained Image-Editing Diffusion Models (ICLR 2024 · similarity 0.86)
- One-Step Diffusion Policy: Fast Visuomotor Policies via Diffusion Distillation (ICML 2025 · similarity 0.85)
- Improving Generative Behavior Cloning via Self-Guidance and Adaptive Chunking (NeurIPS 2025 · similarity 0.85)
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.
