Research map / Video / Video generation
RoboMirror: Understand Before You Imitate for Video to Humanoid Locomotion
Zhe Li, Boan Zhu, Yangyang Wei, Shuanghao Bai, Yuheng Ji, Tao Huang, Pengwei Wang, Zhongyuan Wang, Gary Chan, Chang Xu, Cheng Chi, Jianfei Yang and 1 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 |
|---|---|---|---|
| reconstruct motion latents from VLM latents instead of aligning them with InfoNCE (ablation) | RoboMirror · Nymeria | R@3 (%): 51.1 -> 64.6 | Improved |
| reconstruct instead of align (ablation) | RoboMirror · Nymeria | FID: 62.42 -> 42.42 | Improved |
| reconstruct instead of align (ablation) | RoboMirror · Motion-X | R@3 (%): 56.41 -> 77.66 | Improved |
| reconstruct instead of align (ablation) | RoboMirror · Motion-X | tracking success rate: 0.92 -> 0.95 | Improved |
| use Qwen3-VL instead of Qwen-VL (ablation) | RoboMirror · Nymeria | R@3 (%): 61.1 -> 64.6 | Improved |
| use Qwen3-VL instead of Qwen2.5-VL (ablation) | RoboMirror · Motion-X | R@3 (%): 77.58 -> 77.66 | No clear effect |
| use Qwen3-VL instead of Qwen-VL (ablation) | RoboMirror · Nymeria | success rate: 0.97 -> 0.99 | Improved |
| use latent-driven instead of pose-driven tracking (ablation) | RoboMirror · Motion-X (third-person) | time (s): 9.22 -> 1.84 | Improved |
| use latent-driven instead of pose-driven tracking (ablation) | RoboMirror · Motion-X (third-person) | success rate: 0.94 -> 0.95 | No clear effect |
| use diffusion reconstruction with kinematic motion latent instead of fine-tuned VLM baseline | RoboMirror · Nymeria tracking (IsaacGym) | success rate: 0.92 -> 0.99 | Improved |
| use diffusion reconstruction instead of fine-tuned VLM baseline | RoboMirror · Nymeria tracking (MuJoCo) | success rate: 0.69 -> 0.78 | Improved |
| use RoboMirror instead of Vid2Mot | RoboMirror · Nymeria | MT-R@3: 49.34 -> 64.60 | Improved |
Closest papers by meaning
- SUGAR: A Scalable Human-Video-Driven Generalizable Humanoid Loco-Manipulation Learning Framework (NeurIPS 2026 · similarity 0.89)
- Robots Pre-train Robots: Manipulation-Centric Robotic Representation from Large-Scale Robot Datasets (ICLR 2025 · similarity 0.88)
- DiT4DiT: Jointly Modeling Video Dynamics and Actions for Generalizable Robot Control (NeurIPS 2026 · similarity 0.88)
- DynaFLIP: Rethinking Robotics Perception via Tri-Modal-Dynamics Guided Representation (NeurIPS 2026 · similarity 0.87)
- RoboScape: Physics-informed Embodied World Model (NeurIPS 2025 · similarity 0.87)
- Universal Humanoid Motion Representations for Physics-Based Control (ICLR 2024 · similarity 0.87)
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.
