According to @berkeley_ai, researchers have developed a method to massively scale robot datasets using just a smartphone scan and a single human hand demonstration video. This innovation eliminates the need for traditional teleoperation, dynamic simulation, or physical robot hardware, enabling the rapid generation of thousands of diverse robot trajectories. These datasets are directly compatible with diffusion policy and VLA models for advanced robotics training (Source: @letian_fu on Twitter, May 16, 2025). This breakthrough could significantly accelerate robotics AI development, potentially driving demand for AI infrastructure tokens and robotics-related cryptocurrencies as the industry seeks scalable training solutions.