Research map / Image generation / Fast diffusion sampling
NaP-Control: Navigating Diffusion Prior for Versatile and Fast Character Control
Chia-Wen Chen, Yan Wu, Korrawe Karunratanakul, Siyu Tang
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 |
|---|---|---|---|
| use RL-learned noise instead of test-time guidance | NaP-Control vs UniPhys · Far goal reaching (flat) | success rate: 81.9% -> 98.4% | Improved |
| predict noise directly instead of iterative guidance | NaP-Control vs UniPhys · Far goal reaching (flat) | FPS: 2.9 -> 22.5 | Improved |
| predict noise directly instead of iterative guidance | NaP-Control vs UniPhys · Velocity control (flat) | FPS: 5.2 -> 24.4 | Improved |
| diffusion-prior noise navigation instead of VAE-prior PULSE | NaP-Control · Agile hand reaching (flat) | jerk: 3631 -> 1146 | Improved |
| diffusion-prior noise navigation instead of PULSE | NaP-Control · Far goal reaching (flat) | success rate: 0.998 -> 0.984 | No clear effect |
| noise navigation instead of UniPhys guidance | NaP-Control · Velocity control (flat) | velocity error: 0.7971 -> 0.1753 | Improved |
| noise navigation instead of UniPhys guidance | NaP-Control · Velocity control (flat) | jerk: 401 -> 752 | Hurt |
| noise navigation instead of PULSE on uneven terrain | NaP-Control · Far goal reaching (terrain) | success rate: 0.966 -> 0.860 | Hurt |
| use compact state representation instead of full UniPhys state (ablation) | diffusion prior · Far goal reaching (flat) | exploration efficiency | Improved |
| use compact state instead of root-only state (ablation) | diffusion prior · Far goal reaching (flat) | performance | Improved |
| navigate both state and action noise instead of action noise only (ablation) | NaP-Control · Agile hand reaching (flat) | performance | Improved |
| use smaller action chunk size (ablation) | NaP-Control · Uneven terrain and agile hand reaching | stability and precision | 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.
- PULSE — PULSE: Self-Supervised Photo Upsampling via Latent Space Exploration of Generative Models (CVPR 2020) · better on 5 of 11 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 0.87)
- One-Step Diffusion Policy: Fast Visuomotor Policies via Diffusion Distillation (ICML 2025 · similarity 0.87)
- Enhanced Fine-Grained Motion Diffusion for Text-Driven Human Motion Synthesis (AAAI 2024 · similarity 0.87)
- Two-Steps Diffusion Policy for Robotic Manipulation via Genetic Denoising (NeurIPS 2025 · similarity 0.86)
- Human Motion Diffusion as a Generative Prior (ICLR 2024 · similarity 0.86)
- Optimizing Diffusion Noise Can Serve As Universal Motion Priors (CVPR 2024 · similarity 0.86)
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.
