# GazeAct Uses Human Eye Tracking to Train Robot Policies

> A dataset pairs robot demonstrations with synchronized human gaze. Its authors report higher task success when gaze guides training, while robots run without gaze input.

Oossa · 2026-10-20 · https://oossa.com/en/gazeact-uses-human-eye-tracking-to-train-robot-policies

Researchers introduced GazeAct, a real-robot dataset and training method that pairs 12,000 demonstrations across 40 tasks with human gaze recorded at 300 Hz. The gaze helps train vision-language-action policies to focus on relevant objects and locations; the authors say the robots do not need gaze input when running.

Across 18 real-world tasks, the authors report average success rising from 48.1% to 58.6%, with gains of up to 25 percentage points. They also report an increase on the LIBERO-Spatial benchmark, from 57.9% to 69.6%.

## The facts

- 12,000 demonstrations across 40 tasks and five domains, with gaze synchronized at 300 Hz.
- Reported average success across 18 real-world tasks: 48.1% to 58.6%.

## Why it matters

For robot researchers, GazeAct offers a way to use operator eye movements as training guidance without requiring gaze sensors during deployment; whether the gains hold beyond the reported tests remains unclear.

## Sources & references

1. [GazeAct: A Human-Gaze-Annotated Manipulation Dataset for Attention-Regularized Policies](https://arxiv.org/abs/2610.24507) – arXiv cs.RO, 2026-10-20

Last updated: 2026-10-20