Generalist co-founder Pete Florence details performance benchmarks and data efficiency for the company's Gen-1 foundation model.
Insight
Florence: Scaling physical interaction data must precede robotic model architectures
“You need to have data to learn stuff, and everybody's robots were just sitting still, and it just felt like we have to get on this path where like, we're actually moving and physically interacting with the world at scale, and then we'll figure out all the rest…”
Insight
Florence: Simple robotic grippers paired with high intelligence can accomplish most tasks
“Pairing human-level intelligence with even very simple you know, as you said, Vivek end-effectors, like, you can really accomplish a lot in the world.”
Insight
Florence: Model intelligence, not mechanical ability, has always blocked robotics progress
“The main blocker the entire time has been the intelligence, not so much like the sort of mechanical ability.”
Assertion Not checkable as stated
Florence: Generalist's robotics model demonstrates emergent ambidexterity
“One that we continue to see which has been quite surprising, is that we can take the model trained on everything, the raw pre-trained model, and then we can train it on a new task, and for that new task we might only ever use the right hand when we are demonst…”
Assertion Not checkable as stated
Florence: Generalist robots generalize to untrained tools for complex tasks
“Another one is the ability to have some, like, ingenuity around how to use tools in a way that was also not trained for the task. So basically we can ask the robot to do a certain type of task where we've only trained it with one type of tool. We can give it a…”
Assertion Not checkable as stated
Florence: Generalist can train robot models on new tasks in minutes
“Yes, we are now starting to get into the modes where we can train a model on, on a new task in, in minutes.”