Learning Library

Full Analysis

# Visual Identity Prompted Multi‑View Video Augmentation for Robotics **Authors:** Boyang Wang, **Source:** [HuggingFace](https://huggingface.co/papers/2601.05241) | [arXiv](https://arxiv.org/abs/2601.05241) **Published:** 2026-01-09 **Organization:** Hugging Face ## Summary - Introducing “visual identity prompting” supplies diffusion models with explicit object cues, enabling generation of consistent multi‑view videos that preserve object appearance across frames. - The generated videos serve as high‑fidelity data augmentations, enriching the visual diversity of manipulation datasets without manual collection. - Training robot policies on this augmented data yields measurable gains in success rates and robustness, both in simulation and on real‑world robot platforms. - The approach is model‑agnostic: any diffusion‑based video generator can be guided by identity prompts, making it easy to integrate with existing simulation pipelines. ## Abstract Visual identity prompting enhances manipulation data augmentation for robot policies by providing explicit visual guidance to diffusion models, improving policy performance in both simulation and real-world settings. --- *Topics: robotics, computer-vision, reinforcement-learning* *Difficulty: advanced* *Upvotes: 19*