Generalist AI has launched a new robotic model, GEN-1.5, that learns tasks just by watching humans. It can replicate complex physical movements after viewing a demonstration that lasts just 3 to 12 seconds. 🤖
This is what developers call "physical prompting." It works like a shortcut: instead of writing custom code for every specific movement, you simply show the robot what needs to be done. It picks up the core objective from the video rather than just memorizing a rigid sequence.
The model can even interpret how human hands perform a task and translate those movements directly to its own mechanical grippers without needing to be pre-trained on simulators.
Previously, getting a robot to handle even a simple chore required massive datasets and tedious, custom fine-tuning for every single action. Now, Generalist AI is replacing that rigid, programmed setup with direct observation of the real world.
However, the technology is still a work in progress. In trials with 10 different tasks, the model showed a 59% success rate, meaning it failed to complete the job in four out of every ten attempts. 📉
Robots are still finding their footing in our messy, physical world.
#robotics #AI #technology
Xataka IA
Subscribe
