According to ABHINAV GUPTA, a researcher manager of Facebook (translated by the former company name: the new company name is META), and Tess Helleekers, the purpose of RESKIN, the purpose of RESKIN is a variety of touch -based.Providing the source of contact data that is useful to evolve AI in the task.
GUPTA and Hellebrekers say that AI models, which have a tactile sensory skill developed using Reskin, may be able to play an active part in healthcare and grasp soft objects.In addition, RESKIN can collect visual, hearing, and tactile data by integrating with other sensors, and creates multimodal datasets, making it more physically realistic world model than before.It is also useful for building.
"Our tactile sensation helps navigate the surrounding world. When we wear shoes, such as light, heavy, soft, hard, stable or stable.You can gather information about the objects you use on a daily basis, such as when preparing for meals, and when you are preparing for meals. AI today effectively incorporates sensations such as visual and hearing, but tactile sensation is still a challenge.This is one of the causes of the limited access to tactile data in nature. As a result, AI researchers who try to incorporate touch -sensation into models are as rich as humans.I'm having a hard time using redundancy. "
Reskin, a deformed elastomer embedded with magnetic particles, has $ 15 with 1,000 DIGIT production, while 100 productions, less than $ 6 per dollar.GUPTA and Hellebrekers say that the thickness is 2-3mm for Digit's 18mm, which can handle more than 50,000 interactions.
"In addition, RESKIN can provide high -frequency three -axis touch signals, so it can also support high -speed operation tasks such as slipping, throwing, catching, and hitting. When exhausted, it is easily peeled off and new.You can exchange it. "(From the blog).
Neither Digit nor RESKIN is the first touch sensor for this type.In addition, the robot gripper called GELFLEX, developed by OMNITACT and MIT's computer science and artificial intelligence institute, realizes a human -like sense of human -like nuances with LEDs and two cameras.Singapore National University has also developed a robot with a touch sensor using Intel's prototype chips.
However, what is called "soft skin" is said to be difficult to mass production due to the variation in use.In order to confirm the reaction of each sensor, each sensor must take the calibration procedure.Furthermore, the material of the sensor changes over time and changes depending on the usage, so the calibration must be correspond to the change.
RESKIN does not need to electrically connect soft materials and electronic devices for measurement, so they can overcome these hurdles.Since the sensor magnetic signal depends on the proximity, the electronic device should be close even if it is not connected.In addition, RESKIN uses a mapping function learned based on data obtained from multiple sources, which is more versatile than conventional mapping functions and achieves "solid" mapping.In addition, the sensor automatically performs fine -tuning (continuously) using a small amount of label -free data using a model without a teacher.
In less teachers, algorithms are applied to "unknown" data without prior categories and labels.On the other hand, in a teacher with a teacher, he learns algorithms using the input data that has been noted to obtain a specific output, and learns until the fundamental relationship can be detected.A mechanical learning system without teachers, such as RESKIN, needs to learn to classify data without labeling, and learns from the structure of data, not from annotations.
"In machine learning systems such as RESKIN, data must be processed so that it is learned from the data structure instead of annotation. For example, two of the two contact points that are physically close,We know that a more similar tactile signal is obtained. Eskin expands the possibilities of diverse and inexpensive tactile modules that cannot be realized with existing systems. The existing camera -based touch sensor isIt requires the minimum distance between the surface and the camera, so the design will be larger. Reskin can be installed on the hands and arms of humans and robots. "GUPTA and Hellebrekers).
Facebook researchers performed some experiments to demonstrate the usefulness of RESKIN, indicated that it could be applied to operations in the hand, identifying contact position, and other power measurement tasks.For example, a RESKIN and a circuit board is installed on the dog's soles, and when the dog is resting, walking, or running, the magnitude and direction of the power is tracked.
"Low cost, compact, and long -life RESKINs were born from research on common tactile sensors. You can easily replace the skin so that the plaster is peeled off, so you can use it immediately, and the model you learned immediately.I think it will be a powerful tool that helps researchers to build AI models that demonstrate various applications abroad. ")
(Continue next)
[Via VentureBeat] @Venturebeat
[Original text]
BRIDGEでは会員制度「BRIDGE Members」を運営しています。会員向けコミュニティ「BRIDGE Tokyo」ではテックニュースやトレンド情報のまとめ、Discord、イベントなどを通じて、スタートアップと読者のみなさんが繋がる場所を提供いたします。登録は無料です。