# ATLAS: robots that learn the choices operators make

ATLAS is a research project of TOS Robotics and Eindhoven University of Technology. It develops an AI method that lets industrial robots learn complex tasks by watching skilled operators, including the choices those operators make. ATLAS stands for Advanced Teleoperation Learning for Autonomous Systems.

Source: https://tosrobotics.com/research/atlas/
Last updated: 2026-11-13

- Partners: TOS Robotics and TU Eindhoven
- Runs: May 2026 to April 2028
- Level: Lab prototype, TRL 4
- Results: Published open access

Existing imitation learning copies the motion of a demonstration and ignores the decisions behind it. **Which object first. In what order. Along which route.** ATLAS teaches the robot those choices too.

Robot cells in small and medium sized companies are expensive and rigid. Every product change, new layout or small series asks for custom programming, recalibration and scarce specialists. Two thirds of Dutch entrepreneurs report ongoing labour shortages.

## Two building blocks ATLAS adds

- **Choice awareness.** The robot learns which object to grasp first, which sequence to follow and which path to choose when several valid options exist.

- **Self-calibrating sensor fusion.** Several cameras and depth sensors are combined in a plug and play way. The setup stays robust when something is hidden or a camera moves a little, without manual recalibration.

## Who does what

Operators demonstrate tasks by teleoperation with the [Teachbot](https://tosrobotics.com/products/teachbot/). The system imitates and generalises. No reprogramming is needed when conditions change.

**TOS Robotics** provides the teleoperation platform, the data and the industrial integration. **Eindhoven University of Technology** contributes machine learning research and scientific expertise.

The approach is validated in a lab prototype at technology readiness level 4, aimed at food, manufacturing, logistics and greenhouse horticulture.

## The impact

ATLAS opens a realistic path to affordable, adaptive robot cells for companies that depend on manual labour today. It cuts engineering effort and changeover times, raises uptime and moves people from repetitive physical work to quality control and supervision.

The results are delivered as reusable building blocks, with code and reference configurations, and as an open benchmark protocol with evaluation scripts. Scientific results and the benchmark are published open access.

Read how the current technology works on the [capabilities page](https://tosrobotics.com/capabilities/).

This activity is co-financed with the PPS-i funds of the Dutch Ministry of Economic Affairs through TKI ICT. TKI ICT is the top consortium for knowledge and innovation of the ICT sector.
