Physical Prompting

Physical Prompting is a way of getting a robot to do a task by simply showing it once, rather than instructing it with words or code. It's like teaching someone new to cooking by demonstrating a recipe once in person and saying "do it like this," instead of writing out the steps.

What makes this interesting is that none of the robot's internal knowledge gets changed. If you feed the robot a short clip of a person demonstrating the task, alongside what it's currently seeing through its camera, into a short-term "memory window," the robot performs the task as if imitating what it just watched. There's no retraining process that adjusts the robot's decision-making, and no need to write a new program for each task.

This ability isn't something the robotics company designed as a goal from the start — it's said to be a property that emerged naturally after the robot was trained on large amounts of diverse physical human-robot interaction data over a long period. It's similar to how a person who accumulates varied experiences in childhood ends up developing a rough sense of how to handle unfamiliar situations later on.

The article introduces this as "an approach called 'Physical Prompting.'" It cites the robot model GEN-1.5, which reportedly followed a task after watching just a 12-second demonstration video, with no separate training involved. A common misunderstanding: this isn't a process of retraining the robot — it's closer to briefly showing an already-trained robot a reference video.