Startups tag robotic training data with brain waves to solve the physical ai bottleneck

encord collaborates with german neuroscience startup zander labs to manufacture physical training data by recording human operators' brain waves and egocentric video during tasks like playing jenga. while text-based models leveraged the internet, physical ai faces a massive data bottleneck requiring scales up to five times the youtube video corpus. by tagging data with mental states like intent and error, encord aims to improve model training efficiency and overcome the limitations of remote operation and synthetic generation. if successful, this multimodal data creation strategy will accelerate the deployment of capable humanoid robots in industrial environments.