tech
Are brain waves the next unlock for physical AI?
Forget YouTube videos—frontier physical AI models need multiple camera angles, dense annotation, and soon, brain wave readings.
TL;DR
- Encord is developing data tooling to train AI models, specifically focusing on generating real-world physical data for robotics.
- The company is experimenting with brain wave sensors from Zander Labs to capture mental states like error and intent, aiming to create more useful training data.
- The scarcity of physical training data is a major bottleneck for humanoid and warehouse robotics, unlike the abundance of text data available for LLMs.
- Encord collects 'egocentric' video data from workers wearing cameras and data from remotely operated robots, along with experimenting with new modalities like brain waves and arm sensors.
- Dense annotation of physical actions is considered much more valuable for training specific robotic tasks than raw video data, despite higher production costs.
- The economics of generating physical training data are costly, contrasting with the low cost of scraping text data for LLMs.