GEN-1.5: Generalist AI teaches robots new tasks from a single demo
decoder ·
Robotics startup Generalist AI introduced GEN-1.5, an AI model that enables robots to learn new tasks from a single demonstration. A short 3- to 12-second demo is loaded into the model's context window as a 'physical prompt,' allowing the robot to perform the task without any training. In ten tests, such as opening a jar or pulling money from a wallet, the model achieved an average success rate of 59 percent, which rose to 83 percent after ten training steps on five minutes of data. The model can chain two prompts into longer sequences, use simulation demos, and partially imitate human hand movements, capabilities that reportedly emerged naturally during over eight months of pretraining on interaction data. Generalist claims to be the first to make in-context learning work across a wide range of tasks, whereas previous research only demonstrated it for a few task types. However, the tasks shown are simple and short, and the results come solely from the company without independent verification.
decoder