Home PC News National University of Singapore used Intel neuromorphic chip to develop touch-sensing robotic...

National University of Singapore used Intel neuromorphic chip to develop touch-sensing robotic ‘skin’

Watch VB Transform’s Conversational AI Summit dwell at the moment!


During the nearly held Robotics: Science and Systems 2020 convention this week, scientists affiliated with the National University of Singapore (NUS) offered analysis that mixes robotic imaginative and prescient and contact sensing with Intel-designed neuromorphic processors. The researchers declare the “electronic skin” — dubbed Asynchronous Coded Electronic Skin (ACES) — can detect touches greater than 1,000 occasions sooner than the human nervous system and determine the form, texture, and hardness of objects inside 10 milliseconds. At the identical time, ACES is designed to be modular and extremely strong to break, making certain it could proceed functioning so long as at the very least one sensor stays.

The human sense of contact is fine-grained sufficient to tell apart between surfaces that differ by solely a single layer of molecules, but nearly all of at the moment’s autonomous robots function solely by way of visible, spatial, and inertial processing strategies. Bringing humanlike contact to machines may considerably enhance their utility and even result in new use circumstances. For instance, robotic arms with synthetic “skin” may make use of tactile sensing to detect and grip unfamiliar objects with simply the correct quantity of stress.

Drawing inspiration from the human sensory nervous system, the NUS staff spent a 12 months and a half growing their system. ACES includes {an electrical} conductor related to a community of sensors, which gather indicators to allow the system to distinguish contact between sensors. ACES takes lower than 60 nanoseconds to detect contact — reportedly the quickest charge up to now for “electronic skin.”

An Intel Loihi neuromorphic chip processes the info collected by the ACES sensors. (Neuromorphic engineering, also referred to as neuromorphic computing, describes the usage of circuits that mimic the nervous system’s neurobiological architectures.) The 14-nanometer processor, which has a 60-millimeter die dimension and comprises over 2 billion transistors, contains a programmable microcode engine for on-chip coaching of asynchronous spiking neural networks (SNNs). SNNs incorporate time into their working mannequin so the elements of the mannequin don’t course of enter information concurrently, supporting workloads like contact notion that contain self-modifying and event-driven parallel computations.

According to Intel, Loihi processes data as much as 1,000 occasions sooner and 10,000 extra effectively than conventional processors, and it could remedy sure forms of optimization issues with good points in pace and power effectivity better than three orders of magnitude. Moreover, Loihi maintains real-time efficiency outcomes and makes use of solely 30% extra energy when scaled up 50 occasions (whereas conventional {hardware} makes use of 500% extra energy). It additionally consumes roughly 100 occasions much less power than extensively used CPU-run simultaneous location and mapping strategies.

Intel neuromorphic skin

Above: A visualization displaying the ACES sensor suggestions.

Image Credit: Intel

In their preliminary experiment, the NUS researchers used a robotic hand fitted with ACES to learn Braille, passing the tactile information to Loihi by way of the cloud. Loihi achieved over 92% accuracy in classifying the Braille letters whereas utilizing 20 occasions much less energy than a normal classical processor, in accordance with the analysis.

Building on this work, the NUS staff additional improved ACES’ notion capabilities by combining imaginative and prescient and contact information in an SNN. To accomplish that, they tasked a robotic with classifying varied opaque containers containing differing quantities of liquid, utilizing sensory inputs from ACES and recordings from an RGB video digicam. Leveraging the identical tactile and imaginative and prescient sensors, additionally they examined the power of the notion system to determine rotational slip, an essential metric for object greedy.

Once this sensory information had been captured, the staff despatched it to each a graphics card and a Loihi chip to check processing capabilities. The outcomes present that combining imaginative and prescient and contact with an SNN led to 10% better object classification accuracy versus a vision-only system. They additionally show Loihi’s prowess for sensory information processing: The chip was 21% sooner than the best-performing graphics card whereas utilizing 45 occasions much less energy.

ACES could be paired with different artificial “layers” of pores and skin, just like the transparent self-healing sensor skin layer developed by NUS assistant professor Benjamin Tee (a coauthor of the ACES analysis). Potential functions embody catastrophe restoration robots and prosthetic limbs that assist disabled folks restore their sense of contact.

Along with Intel, researchers at IBM, HP, MIT, Purdue, and Stanford hope to leverage neuromorphic computing to develop supercomputers a thousand occasions extra highly effective than any at the moment. Chips like Loihi excel at constraint satisfaction issues, which require evaluating numerous potential options to determine the one or few that fulfill particular constraints. They’ve additionally been proven to quickly determine the shortest paths in graphs and carry out approximate picture searches, in addition to mathematically optimizing particular aims over time in real-world optimization issues.

ACES is among the many first sensible demonstration of the expertise’s capabilities, following Intel analysis displaying neuromorphic chips can be utilized to “teach” an AI mannequin to tell apart between 10 completely different scents.

Most Popular

LinkedIn open-sources GDMix, a framework for training AI personalization models

LinkedIn recently open-sourced GDMix, a framework that makes training AI personalization models ostensibly more efficient and less time-consuming. The Microsoft-owned company says it’s an...

Google’s Smart Cleanup taps AI to streamline data entry

In June, Google unveiled Smart Cleanup, a Google Sheets feature that taps AI to learn patterns and autocomplete data while surfacing formatting suggestions. Now,...

ExamSoft’s remote bar exam sparks privacy and facial recognition concerns

Sometimes the light Kiana Caton is forced to use gives her a headache. On top of common concerns that come with taking a state...

Monetization strategies for the skyrocketing LatAm mobile game market (VB Live)

Presented by Facebook Audience Network There are hurdles to launching a mobile game internationally, but there’s never been a better time to go global. For...

Recent Comments