# RoboRender creates photorealistic robot videos to bridge simulation gap

> Researchers show that policies trained on RoboRender‑generated videos reach 71% success on real robots, far outpacing raw simulation.

Oossa · 2026-10-08 · https://oossa.com/en/roborender-creates-photorealistic-robot-videos-to-bridge-simulation-gap

A team led by Huang Huang introduced RoboRender, a system that turns simulated robot trajectories into realistic RGB videos. The model uses depth videos, language commands, and robot masks to add real‑world textures and backgrounds while keeping the underlying motion and actions. Policies learned from these videos were tested on pick‑and‑place, articulated‑object, and mobile‑manipulation tasks and succeeded 71% of the time. That is about 7.1 times better than training on raw simulation renders and 3.6 times better than standard visual domain randomization.

## The facts

- 71% average success rate on real‑world tasks
- Paper posted on 2026‑10‑08

## Why it matters

Robot developers can train controllers in simulation and deploy them directly, cutting costly real‑world data collection.

## Sources & references

1. [RoboRender: Robot-Oriented Video Generation for Visual Sim-to-Real Transfer](https://arxiv.org/abs/2610.09254) – arXiv cs.RO, 2026-10-08

Last updated: 2026-10-08