Users call the single-pass learning of one shared model the real shift in LingBot VA 2.0's robot AI training.
No Digg Deeper questions have been answered for this story yet.
One system, learned in a single pass. That is the real shift here. @robbyant_brain
LingBot VA 2.0 does it differently. It learns what it sees and what it does in one shared model, at the same time, instead of stacking pieces on top of each other. That is why it can pick up a new task from just 10 to 15 examples in the paper's tests. It learns to focus on the details that actually matter for moving and acting, not just what things look like.
LingBot VA 2.0 does it differently. It learns what it sees and what it does in one shared model, at the same time, instead of stacking pieces on top of each other. That is why it can pick up a new task from just 10 to 15 examples in the paper's tests. It learns to focus on the details that actually matter for moving and acting, not just what things look like.
Most robot AI is built like a stack of separate parts. A generic video model to see, a bolted on piece to track what is happening in the world, and another piece underneath to decide what to do. Three systems duct taped together. @robbyant_brain
Project page: https://technology.robbyant.com/lingbot-va-v2 Paper: https://github.com/Robbyant/lingbot-va/blob/main/LingBot_VA2_paper.pdf
One system, learned in a single pass. That is the real shift here. @robbyant_brain
