Facebook's parent company Meta Platforms Inc. is pushing the boundaries of artificial intelligence robots to touch sensitivity through its two new sensors.
They include a high-resolution robotic fingertip sensor called DIGIT (pictured) and a thin and replaceable robotic "skin" called ReSkin, which can help artificial intelligence robots recognize the texture, weight, temperature, and state of objects, etc. information.
According to Meta AI researchers Roberto Calandra and Mike Lambeta, the idea is to use so-called "tactile perception" to train robots to extract key information from the things they touch, and then combine these proprietary technologies with other information to perform tasks. Greater complexity.
Tactile sensing aims to replicate human-level touch in robots so that artificial intelligence can "learn and use touch itself, as well as combine with other sensing methods such as vision and audio." Another hypothetical benefit is that, for example, robots with tactile sensing capabilities will be gentler and safer when handling other objects or objects.
In order for artificial intelligence robots to use tactile data and learn from it, they first need to be equipped with sensors that can collect information from the things they touch.
"Ideally, touch-sensing hardware should simulate many of the properties of a human finger," Calandra and Lambeta said. "They should be able to withstand the abrasion caused by repeated contact with the surface... [they] also need to have high resolution and be able to measure a wealth of information about the object being touched, such as surface characteristics, contact force, and other characteristics that can be distinguished by contact. "
Meta open sourced the blueprint for this sensor DIGIT in 2020, saying at the time that it was easy to build, reliable, low-cost, compact and high-resolution. The design has been widely adopted by universities and research laboratories, and now Meta hopes to take DIGIT to a new level by collaborating with a start-up company called GelSight Inc. to commercially manufacture sensors and make them more widely available for research. People use and accelerate innovation.
Meta explained that for ReSkin, it is a brand new sensor, just like the artificial skin of a robot hand.
Meta AI researchers Abhinav Gupta and Tess Hellebrekers wrote in a blog post. "The AI model trained through the learned tactile perception skills will be able to perform many types of tasks, including tasks that require higher sensitivity, such as working in a medical environment, or require greater dexterity, such as manipulation of small, soft, or sensitive Objects."
The advantage of ReSkin lies in its extremely low production cost. Meta claims to be able to manufacture 100 units for less than $6, or even cheaper in mass production. ReSkin is 2 to 3 mm thick, has a life span of 50,000 interactions, has a high temporal resolution of up to 400 Hz and a mm spatial resolution of 90% accuracy.
Such specifications make it very suitable for many purposes, such as manipulators, tactile gloves, arm covers and even dog shoes. Therefore, it should enable researchers to collect a large number of different types of tactile data that were impossible to collect before, or at least very difficult and expensive. In addition, ReSkin also provides high-frequency, three-axis tactile signals, which can realize fast operation tasks, such as throwing, slipping, catching and clapping.
In order to further help the tactile sensing research community, Meta created and open sourced a simulator called TACTO, which can be experimented without hardware. Simulators are the key to advancing AI research in most areas because they enable researchers to test and verify hypotheses without the need for time-consuming experiments in the real world.
The idea of TACTO is to help researchers simulate vision-based tactile sensors with different shapes installed on different types of robots.
"TACTO is capable of rendering realistic high-resolution touch readings at hundreds of frames per second, and can be easily configured to simulate different vision-based tactile sensors, including DIGIT," Calandra and Lambeta said.
It is one thing to have all the tactile sensor data, but researchers also need a way to process it and gain insights. Meta, which was once helpful, helped this field by creating a library of machine learning models called PyTouch, which can convert raw sensor readings into advanced properties, such as detecting sliding or identifying materials that have been touched by the sensor.
Researchers can use PyTouch to train and deploy various models across different sensors. These models provide basic functions such as detecting touches, sliding, and estimating object poses.
Calandra and Lamberta said.
Ultimately, it is hoped that PyTouch will help researchers use advanced machine learning models dedicated to touch sensing "as a service". The idea is that researchers will be able to connect a DIGIT sensor, download a pre-trained model, and then use it as a basic building block for their robotic application.
Meta AI said that despite all the progress made in tactile perception so far, there is still a lot of work to be done. Calandra and Lambeta explained that in order to develop a robot with true human-like touch sensitivity, more hardware is needed—for example, a sensor that can detect the temperature of an object. They also need to better understand which touch functions are most important for a particular task, and gain insight into the correct machine learning calculation structure for processing touch information.
The researchers wrote: "Improvements in touch sensing can help us advance artificial intelligence and enable researchers to build robots with enhanced functions and capabilities." "It can also open up the possibilities of AR/VR and lead the industry and medical And the innovation of agricultural robot technology. We are working hard to realize a future in which every robot may be equipped with touch-sensing functions."
Click here to join the free and open Startup Showcase event.
We really want to hear from you, and we look forward to seeing you at the event and the CUBE club.
Click here to join the free and open Startup Showcase event.
US states investigate Instagram’s harm to young people
VMware brings its cross-cloud service for multi-cloud application development to AWS
The social security number in the California pizza kitchen data breach was stolen
With strong earnings, Palo Alto Networks provides bullish guidance for the rest of the year
Comet, an enterprise machine learning development platform, raises US$50 million
Workday acquired VNDLY, a contractor management platform, with strong earnings but falling stock price
US states investigate Instagram’s harm to young people
Policy-James Farrell. 5 hours ago
VMware brings its cross-cloud service for multi-cloud application development to AWS
Cloud-Author: Mike Whitley. 6 hours ago
The social security number in the California pizza kitchen data breach was stolen
Security-by Duncan Riley. 7 hours ago
With strong earnings, Palo Alto Networks provides bullish guidance for the rest of the year
Security-by Mike Whitley. 7 hours ago
Comet, an enterprise machine learning development platform, raises US$50 million
AI-Author: Duncan RILEY. 7 hours ago
Workday acquired VNDLY, a contractor management platform, with strong earnings but falling stock price
Cloud-Author: Mike Whitley. 8 hours ago
Like free content? Subscribe to follow.