We Hired 10 Humans to Train a Robot to Fold Shirts

2026-04-09 · Source: HuggingFace · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Advanced, quick

Summary

A fully autonomous robotic system is being developed to fold t-shirts of various sizes and colors, even when initially crumpled. This project, initiated last year, aims to automate complex tasks using open-source tools. The development process involves collecting data, training models, and conducting evaluations. A significant part of the data collection involved hiring approximately 10 teleoperators who worked in pairs to demonstrate t-shirt folding, generating the necessary data for model training. This foundational work is intended to enable the robot to learn to fold other types of clothing as well, demonstrating end-to-end automation of a challenging task.

Key takeaway

For research scientists developing robotic manipulation systems, this project demonstrates a viable path for automating intricate tasks like garment folding. You should consider teleoperation as a scalable method for generating diverse training data, especially when starting from scratch on a complex, unstructured problem. This approach can significantly accelerate model learning and system development.

Key insights

Robots can learn complex manipulation tasks like t-shirt folding through teleoperated data collection.

Principles

• Complex tasks can be automated end-to-end.
• Teleoperation accelerates data collection for robotics.

Method

The method involves collecting teleoperated demonstrations, training models to reproduce folds, and enabling the robot to learn autonomous folding.

In practice

• Use teleoperators for complex data collection.
• Start with flattened items, then crumpled.

Topics

• Robot T-shirt Folding
• Autonomous Robotics
• Teleoperator Data Collection
• Complex Task Automation
• Open-Source Robotics

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Robotics Engineer

Related on AIssential

• Operating a Humanoid With Your Body Is a Hot Job in China’s Hardware Capital — Human teleoperation of humanoid robots collects training data for eventual autonomous functionality in workplaces.
• Dynamic robotic cloth folding with efficient Koopman operator-based model predictive control — Koopman operator regression enables efficient, accurate robotic cloth folding via linearizing complex cloth dynamics.
• Video Friday: Digit Learns to Dance—Virtually Overnight — Advanced robotics are achieving new performance thresholds in control, learning, and human interaction.
• From Web Video to Real-World Robots — Web-scale video pre-training combined with minimal teleoperation data enables highly efficient robot foundation models.
• This startup wants to dismantle EVs to solve Europe’s critical materials problem — Automated EV dismantling using AI and robotics can secure critical material supply chains and meet recycling targets.
• Nvidia research shows robots that train themselves through AI coding agents — AI coding agents empower robots to autonomously achieve up to 99% success in real-world dexterous grasping tasks.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by HuggingFace.