
Gave a talk about the huge question of the 'robot data gap' in training robot foundation models.
How media typically covers Ken Goldberg
Directly quoted in these articles
UC Berkeley roboticist Ken Goldberg argues that a "100,000-year data gap" will prevent humanoid robots from gaining real-world skills like those claimed by tech leaders, debunking near-term timelines for surgical robots or in-home butlers.
“UC Berkeley roboticist who published two papers in Science Robotics explaining why robots are not gaining real-world skills as quickly as AI chatbots.”
Referenced in coverage
Training a robot foundation model equivalent to language model scale (2 trillion tokens) requires 70,000+ robot-years of data, but scaling fleets, simulation, and human video data combined could make this feasible with substantial investment.
“Gave a talk about the huge question of the 'robot data gap' in training robot foundation models.”