NTH

RoboTryOn: Synthesizing Dexterous Robot Demonstrations from Human Videos via Virtual Try-On

AuthorsSeungyong Lee, Suhyeok Jang, Minyoung Kim, Jinwoo Shin

AffiliationsKAIST RLWRLD Korea University Robot foundation models have shown remarkable performance in robot control, yet adapting them to a new task still relies heavily on teleoperated demonstrations. Egocentri

November 6, 2026 2 min read
Watch on YouTube
The one-line take

RoboTryOn transforms human videos into robot demonstrations, aiming to teach robots new tasks without collecting task-specific teleoperation data.

Key results

60.0%
Out-of-domain success

Average success rises from 2.1% without RoboTryOn to 60.0% with synthesized demonstrations.

45.9%
Bimanual task success

Average success on challenging bimanual tasks, despite teleoperated demonstrations being single-arm.

33.3%
Collection cost

Human-video training matches teleoperation performance at roughly one-third the collection cost.

69.7%
Open-loop replay success

RoboTryOn action labels, compared with 34.8% for pose-retargeted actions.

32.7
Fréchet SigLIP distance

Distance to real robot footage, reduced from 125.5.

What the paper found

RoboTryOn turns an egocentric human manipulation video into a robot demonstration by swapping the human embodiment while preserving the scene and motion. It translates selected frames into robot keyframes, synthesizes the motion between them using hand-skeleton flow, then uses an inverse dynamics model to infer robot actions from the generated video instead of directly retargeting human joints. The pipeline uses FLUX.2 [klein] for keyframes, LTX-2.5 for video synthesis, Depth Anything V2 for action labeling, and fine-tunes π0.5 policies; its robot-video training also includes NVIDIA GR00T teleoperation data. Across two humanoid embodiments, average success on tasks without task-specific robot demonstrations rises from 2.1% to 60.0%, and success on challenging bimanual tasks reaches 45.9% despite teleoperated training demonstrations being single-arm. The generated action labels achieve 69.7% open-loop replay success in simulation, compared with 34.8% for pose retargeting, while the visual Fréchet SigLIP distance to real robot footage falls from 125.5 to 32.7. For in-domain tasks, human-video training matches teleoperation performance at roughly 33.3% of the collection cost.

Original abstract

Robot foundation models have shown remarkable performance in robot control, yet adapting them to a new task still relies heavily on teleoperated demonstrations. Egocentric human videos offer a scalable substitute, but overlaying rendered robots on human videos or retargeting human poses into robot actions leaves significant visual or action gaps for high-DoF robot embodiments. To address this limitation, we propose RoboTryOn, which treats human-to-robot translation as virtual try-on: try-on keeps the person and swaps the garment, RoboTryOn keeps the demonstrated motion and swaps the human for the robot. Specifically, RoboTryOn converts a human video into a video of the target robot specified by a reference image, while preserving the original task, objects, and background. To maintain visual consistency, we translate selected keyframes autoregressively and use them to guide video synthesis with motion conditioned on the human hand skeleton. Additionally, we use an inverse dynamics model (IDM) that incorporates depth features to infer action labels from synthesized videos for policy learning. Our experiments across multiple humanoid embodiments show that policies trained with our synthetic data improve the average success rate on novel tasks, including challenging bimanual manipulation, from 2% to 60%, without any real robot demonstration of those tasks. Moreover, training with our synthetic data matches the performance of policies trained solely on teleoperated demonstrations at approximately a third of the collection cost, demonstrating the cost efficiency of our approach.

Read the original paper

More in Embodied AI

Browse all 69 papers →
02Embodied Ai

Kimodo-Policy: Learning Kinematic Action Models for Humanoid Loco-Manipulation

Yunheng Wang, Jiaxi Zhang, Wei Lu, Yuetong Fang, Taowen Wang, Yixiao Feng, Shuning Zhang, Junzhe Xu, Yichi Wang, Zizhao Yuan, Kailun Huang, Lusong Li, Mingqian Zhou, Zecui Zeng, Renjing Xu

Kimodo-Policy turns a pretrained whole-body motion generator into a control policy for humanoid robots that walk and manipulate objects.

Read analysis