On the map

At the centre Robot training data is not just recordings of a task: the environment has to be randomised, or the policy only ever learns the one cable it was shown. Holds FLFei-Fei Li Read this korrent →

• Similar wording A robot going down a wrong path is not producing training data; the efficient move is for a human to intervene and show the recovery. Holds CFChelsea Finn
• Similar wording Robot policies already show emergent capability: one transferred a skill from its right hand to its left with no such example anywhere in its training data. Holds CFChelsea Finn
• Similar wording Hand-tuning a robot's data set cannot reach very high reliability, because people get tired; the system has to seek out its own missing data. Holds CFChelsea Finn
• Similar wording RL environments exist to make a model generalise, not to teach it each skill one at a time — exactly as pre-training does. Holds DADario Amodei
• Similar wording It is the diversity of a robot data set, not its size, that produces generalization: dropping the most diverse slice hurts far more than dropping a random fifth. Holds CFChelsea Finn
• Similar wording Robots will learn from deployment more easily than chatbots do, because a physical mistake is obvious the moment it happens. Holds SLSergey Levine
• Similar wording Once a robot is good enough, you can teach it with words instead of with demonstrations, and language becomes a training signal for motor skill. Holds SLSergey Levine
• Similar wording Most future robotics training data will be robots' own autonomous attempts, the way most language-model data is now generated by the model itself. Holds CFChelsea Finn