Encord tests brain wave headsets to help train robots

In short: Some robotics teams are starting to record people’s brain signals, along with video, to create better training data for robots.

What's going on

Encord, a company that makes tools and services for preparing AI training data, is running a trial in a warehouse in San Leandro, California. Workers there act as “robot trainers” by doing hands-on tasks like taking apart a Jenga tower while wearing a headset that records what they see.

In this trial, the headset also measures brain activity, often called brain waves. The headset comes from Zander Labs, a German neuroscience startup. The idea is to use these signals to guess mental states like surprise, intent, or when someone notices an error.

Encord says this is a test run, not a full rollout. The company plans to build an initial set of training examples that include brain wave tags, then see if customer robot systems perform better when trained on it.

The broader trend is that many robot makers are running into a data problem. Chatbots learned from huge amounts of text on the internet, but robots need real-world examples of physical actions. Encord and others are collecting “egocentric” video (first-person video, like a body cam) and also using human-controlled robot arms to record tasks such as pouring coffee, stacking chips, and plugging in cables.

What to watch

If tests show that brain wave signals help robots learn faster or make fewer mistakes, more companies may add extra sensors like brain wave headsets or muscle sensors. The key question is cost. Making this kind of data is more like running a film shoot than downloading web pages, and that makes robot training more expensive and harder to scale.

Source: TechCrunch AI