Deepmind partnered with 33 academic labs unveiled this openx embodiment dataset which has been sourced from 22 robot types and RT1X model
-Attempt to achieve general purpose robotics learning across various robot types
-Unified dataset has a primary objective of solving many problems associated with robotics
-Problem
Robotics is a great specialist, but they are also not great journalists.
Each task, robot type, and environment demands a unique training model making modifications challenging
Solution:
Deep mind approach is to bridge this gap, aiming to combine the collective robotics knowledge to shape a more general purpose robot
-Uses a very comprehensive dataset developed with 20 institutions capturing 500 skills and 150,000 tasks from diverse robotics types
-Various Tasks:
Picking an icecream and identifying the icecream by itself
Moving the object from one place to another
-Dataset:
Has 527 skills
-New RTX Model
Built on their previous robotics transformer model RT1X and RTX2 and has shown improvements due to the diverse data from the new dataset
The model displays adaptability which showcase abilities not actually not initially present in its training set
Footnote:
Embodiment?
Refers to giving a digital AI system a physical form, typically in a robot
This allows the AI to interact directly with the physical world. Unlike traditional AI that operates within computational environments (like playing chess or analyzing data), embodied AI can move, perceive objects, and perform tasks in the real world
This is achieved through the robot’s sensors and actuators:
Sensors:
Allow the robot to gather information from its environment
This can include visual data from cameras (similar to human eyes), tactile feedback from touch sensors (similar to human skin), or auditory data from microphones (similar to human ears)
Actuators:
Enable the robot to perform actions, such as moving its limbs, wheels, or other parts to navigate space, manipulate objects, or interact with its environment in other ways
Reference: https://www.youtube.com/watch?v=umySOgmrPpI
Reference: https://robotics-transformer-x.github.io/