Generalist AI Unveils GEN-1.5: Enabling One-Shot Learning for Robots

Industry Brief Generlaist AI shows “physical prompting” approach to learn news tasks with a handful of demonstrations The News Generalist AI has announced GEN-1.5, an advanced embodied foundation model capable of learning new physical tasks from as little as 3 to 12 seconds of demonstration data. This development signifies a major step towards robots that can acquire new skills with minimal human intervention. What It Is GEN-1.5 is the latest iteration of Generalist AI's embodied foundation models, designed to bring general intelligence to the physical world. It is an end-to-end neural network that allows robots to learn and adapt to new tasks rapidly, either through in-context learning from short demonstrations or by adapting with a few gradient steps on minutes of data. Key Highlights One-Shot Learning: GEN-1.5 can learn new physical tasks from demonstrations as short as 3 to 12 seconds, or adapt with 1 to 10 gradient steps on minutes of data. Pretraining on Physical Experience: These capabilities emerge from extensive pretraining on large-scale physical interaction data. Broad Capabilities: The model demonstrates broad capabilities across one-shot and few-shot learning from demonstration, as well as zero-shot physical generalization. Foundation Model Evolution: GEN-1.5 builds on previous models like GEN-0 and GEN-1, which established scaling laws in robotics and achieved mastery of simple physical tasks with high success rates and speed. Beyond Fixed End Effectors:…