Key facts
- Encord is experimenting with brain wave sensors to collect data for training AI models.
- The company is working with Zander Labs, a German neuroscience startup, to measure brain activity.
- The data aims to capture mental states like error, intent, and surprise to improve AI training.
- This initiative seeks to address the scarcity of real-world physical training data for robotics.
Encord, a company specializing in AI training data, is exploring the use of brain wave sensors to enhance the development of physical AI. The company is conducting trials with Zander Labs, a German neuroscience startup, to measure brain activity during tasks like disassembling a block tower. The goal is to capture mental states such as error, intent, and surprise, which could provide more valuable data for training robotic models than traditional methods.
This initiative addresses a significant challenge in robotics: the scarcity of real-world physical training data. Encord's head of robot learning, Vineeth Velmurugan, stated that the necessary data scale is immense, potentially five times the size of YouTube's video corpus, making data generation a critical business. The company is experimenting with various data modalities, including "egocentric" video collected by workers and electrical signals from forearm sensors, alongside brain waves.
Encord's pilots, such as Andrew Ceja and Sofia Infante, perform tasks like stacking poker chips and manipulating server cables to generate data for specific robotic skills. While dense annotation of this data is costly, it is considered significantly more valuable than raw "ego data." The economics of generating this specialized physical training data differ from the low cost of scraping internet text for large language models, highlighting a key distinction in the development of physical AI.