movie
While AI sometimes shows talent that surpasses humans in difficult tasks, such as beating humans at chess or creating new drugs, it sometimes struggles with tasks that humans can easily handle. This is explained by Hans Moravec, a mechanical engineer.Moravec’s paradox”, but since we also need to work on solving physical problems, we decided to work with a group that develops basic models and learning algorithms.Physical Intelligencetook eight months to develop the general-purpose robot-based model “π0” as an AI model for robots that perform general-purpose movements.
Our First Generalist Policy
https://www.physicalintelligence.company/blog/pi0
According to Physical Intelligence, “π0” is the first step in the long-term goal of developing “Artificial Physical Intelligence”, similar to using large-scale language models (LLM) and chatbot assistants. , you can ask the robot to solve any task.
“π0” learns a variety of data, so it can execute a variety of instructions. Unlike LLM, the robot learns to output direct movement commands through a new architecture through training based not only on images and text, but also on concrete experiences with the robot, acquiring “physical intelligence” I’m doing it.
Demonstration videos show “π0” performing tasks such as taking laundry out of the washing machine and folding it, cleaning the table after a meal, and assembling boxes.
A robot arm opens the lid of a drum-type washing machine and takes out laundry.
How to fold the laundry taken out of the washing machine one after another
How to clean up the dishes left after a meal
How coffee beans are weighed and transferred to the mill
How to assemble the box
The first prototype model was trained on open source datasets and dexterity task datasets from eight different robots. The reason why data containing a wide range of tasks rather than specialized tasks was used as a training source is to teach the underlying model general knowledge of physical interactions rather than solving a specific application. , described by Physical Intelligence as “the first foundation of physical intelligence.”
In order to successfully build the model, Physical Intelligence says it believes it will be necessary to collaborate with the entire robotics community, and has already begun collaborating with a number of companies and research institutes. .
Copy the title and URL of this article
