고급 AI를 통한 촉감 모방 – 가능할까?

내일의 로봇은 인간과 비교할 수 있는 감각을 갖게 될 것입니다. 이미 장치를 시각, 후각, 청각으로 만들 수 있는 AI 알고리즘이 존재합니다. 이러한 향상은 센서와 고급 알고리즘을 활용해 이러한 감각을 시뮬레이션합니다. 그러나 지금까지 연구자들을 피한 감각이 하나 있습니다: 촉감.

현재 스티븐스 공과대학의 연구진이 촉감을 모방하는 새로운 방법을 공개했습니다. 알아야 할 내용은 다음과 같습니다.

연구자들은 오랫동안 로봇 시스템을 통해 촉감을 모방하려고 노력해 왔습니다. 감각을 가진 로봇은 다양한 활용 사례의 문을 열어줄 것입니다. 더 많은 사람들이 이러한 장치와 함께 일하기 시작하면서 ‘감각’을 가진 로봇에 대한 관심이 지속적으로 높아지고 있습니다. 엔지니어들은 이 감각이 효율성, 능력 및 작업장 안전을 향상시키는 데 필수적이라고 믿고 있습니다.

Think of robots that could tell if they bump into you and react accordingly. This ability would allow robots that work closely with humans to reduce their risks and enable them to conduct more precise and previously human-only tasks.

고급 AI 연구를 통한 촉감 모방

The study¹, published in the journal Applied Optics, reveals how researchers were able to simulate touch via lasers and an AI algorithm. The researchers combined AI and quantum technology to achieve this goal in a quantum lab setup.
Quantum interactions provide a plethora of data that can be used to create and improve AI models. This approach combines advanced machine learning, a raster scanning single photon LiDAR, and quantum feedback in the form of speckle noise to accomplish this task.

광자 발사 스캔 레이저를 이용한 촉감 모방

The team created a raster scanning single photon LiDAR that could be set up to pulse at different intervals, enabling engineers to document any changes and feedback reflected from the surface. Each surface has different refractions depending on its makeup. Researchers recognized this fact and determined this was the best way to artificially provide the system with the ability to quickly conduct topography scans.

알고리즘 AI 모델

The AI scans the surface using the proprietary laser as the first step. This maneuver creates a detailed image of the item. The detailed image will produce random feedback called speckle noise. In the past, speckle noise was a hindrance to optical clarity as this feedback lowered resolution.

The researchers noticed that this feedback was more than interference. It provided a unique signature for every surface based on its roughness. This data was fed to the AI system, which then deciphered the information to determine the structure’s dimensions, height, and roughness.

The researchers utilized this setup to scan multiple types of surfaces to program the AI algorithm. Specifically, the system registered backscattered protons from different points across the surface. From there, the data is gathered and formed into a mode fiber that is counted via a single photon detector. This device can differentiate speckle noise from other interference, enabling researchers to leverage this interference to determine the object’s smoothness.

촉감 모방 테스트

To test their theory, the team began by setting up a single pixel, raster scanning single photon counting LiDAR system. This device was optimal for the research as it can produce a collimated laser beam that can be fired in picosecond pulses, providing accurate coverage and responsiveness.
The engineers decided to utilize 31 industrial sandpapers as their test subjects. They started by acquiring all different varieties and roughness. Specifically, the sandpaper ranged from 1 to 100 microns thick. The laser then passed pulses through the transceiver and at the sandpaper. The light and interference were refracted back and calculated by the AI system.

촉감 연구 결과

The results of these tests show promise. The new system initially had an accuracy of 8 microns, which was reduced to 4 microns after fine-tuning and adjustments. Notably, this level of accuracy is comparable to the industry’s leading solutions.
Notably, the system appeared to work best when the surface had a fine grain rather than major roughness. Impressively, the engineers were able to accurately determine the surface structure of the sandpapers with minimal effort, opening the door for this technology to reshape how industries operate in the coming years.

촉감 연구의 이점

This research could unlock many benefits across multiple industries. For one, this method would provide huge cost savings over the current systems in use. Additionally, it would allow manufacturers to reduce their employees as the AI is far more accurate. This maneuver would reduce overhead and improve their bottom line.

빠른 표면 매핑

시스템이 표면을 스캔하는 속도도 또 다른 장점입니다. 이 방법은 레이저가 표면을 몇 순간 스캔하면 판단이 이루어지므로 매우 빠르고 사용에 필요한 노력이 적어 제조업체가 더 많은 스캔을 수행하고 비용을 절감할 수 있습니다.

저비용 통합

이 연구의 또 다른 큰 장점은 시장에 저비용 통합 솔루션을 제공한다는 점입니다. 많은 경우 LiDAR는 핵심 부품의 구조적 무결성을 판단하는 데 사용됩니다. 새로운 시스템은 기존 LiDAR를 향상시켜 마이크론 수준의 측정을 가능하게 할 수 있습니다.

품질 관리

새로운 표면 스캔 방식은 정교하고 정확한 부품의 품질 관리 방법을 개선할 것입니다. 엔지니어들은 오랫동안 시스템을 사용해 항공기 및 기타 부품의 중요한 구성 요소가 심각한 문제로 이어질 수 있는 결함이 없음을 보장해 왔습니다.

향상된 분류

로봇 분류기는 이미 전 세계적으로 높은 수요와 사용을 보이고 있습니다. 이 업그레이드는 제품의 구성 및 필요한 분류를 판단할 때 추가적인 감각을 제공함으로써 시스템의 역량을 향상시킬 수 있습니다. 예를 들어, 촉감을 가진 로봇 손은 농산물의 단단함을 검사해 익었는지 여부를 판단할 수 있습니다.

촉감 연구자들

Engineers from the Stevens Institute of Technology spearheaded the research into the laser topography system. They worked closely with CQSE Director Yuping Huang as part of their project. Additionally,  Daniel Tafone and Luke McEvo received credit for their efforts on the project.

잠재적 응용 분야

There are a lot of applications for this technology. Already, LiDAR plays a vital role in safety standards. This research improves those capabilities significantly and enables engineers to conduct never-before-possible real-time monitoring of crucial components.

헬스케어

헬스케어 산업에서는 인간처럼 촉감을 가진 로봇에 대한 수요가 증가하고 있습니다. 이러한 시스템은 산업 내에서 다양한 용도로 활용될 수 있습니다. 흥미로운 사례 중 하나는 이 장치가 점을 스캔해 치명적인 흑색종을 탐지하도록 하는 것입니다. 레이저 기반 알고리즘은 하나의 점이 안전하고 다른 점이 잠재적으로 치명적인 미세 차이를 판단해 수천 명의 환자 생명을 구할 수 있습니다.

LiDAR 향상

LiDAR는 오늘날 다양한 제품에 사용됩니다. 스마트카, 로봇, 스마트폰 및 기타 제품은 LiDAR를 눈 역할로 활용합니다. 로봇 청소기에도 장애물을 피하기 위한 형태의 LiDAR가 포함되어 있습니다. 이 새로운 기술은 마이크로 로봇이 인체와 같은 환경을 탐색하고 필요한 위치에 직접 생명을 구하는 치료를 전달하도록 도울 수 있습니다.

이 연구로 혜택을 받을 수 있는 기업

Several robotics firms could integrate this technology and improve their results today. Robotics is a fast-growing sector that now spans nearly every industry. From conducting surgeries to picking fruit, these devices could see a major boost with the introduction of a touch emulator.

Samsara 

Samsara (IOT )  is a San Francisco-based IoT firm that seeks to make waves. The company was founded by Sanjit Biswas and John Bicket in 2015 to create, manage, and offer enterprise clientele a solid feature to track and monitor logistics. Today, the company provides a wide range of products to accomplish this task, including AI dashcams, route optimization, equipment tracking, site monitoring, and telematics.

Samsara는 IoT(사물인터넷) 시장의 주요 기업입니다. IoT 장치는 오늘날 사용되는 수백만 개의 스마트 디바이스를 의미합니다. 인터넷 연결, 센서 및 데이터 통신 능력을 갖춘 모든 것이 될 수 있습니다. 현재 IoT 분야는 전 세계적으로 수십억 개의 스마트 디바이스를 포함하고 있습니다.

Samsara는 기업이 이러한 장치를 물류에 통합해 결과, 효율성 및 보안을 향상시킬 수 있게 합니다. IoT 장치는 제품의 상태, 진위, 위치 등 실시간으로 모니터링하는 데 활용될 수 있습니다.

분석가들은 IoT 산업이 확대됨에 따라 Samsara가 성장할 좋은 위치에 있다고 평가합니다. 이 회사는 304억 3천만 달러의 시가총액을 보유하고 있으며 업계 주요 기업들의 지원을 받고 있습니다. 특히, 올해 하버드 대학이 선정한 최고 주식 후보 중 하나로 선정돼 소비자 신뢰를 높였습니다.

인간 감각을 모방하려는 다른 시도

Examining the race to make robots feel reveals some interesting developments. The first thing you notice is that there are two very different approaches to making robots gain the sense of touch. Hardware solutions incorporate devices that can simulate touch by registering pressure and heart, whereas software solutions integrate algorithms that utilize feedback to simulate touch. Here are some other ways in which researchers have found to give robots the ability to feel.

고성능 세라믹

A recent study demonstrates how micro-ceramic particles could be embedded in a flexible skin-like layer to enable the device to register heat and pressure. The tiny ceramic particles provide the perfect way to transfer electrical pulses across a flexible surface.

This research saw engineers develop robot skin tags that could tell if you brushed against and pulled away. They then went on to create a smart prosthetic that could enable the user to feel surfaces and react accordingly. They noted that their robotic skin was capable of registering touch using these pulses down to the finest pressure settings.

인공 신경

Another exciting breakthrough in the field of robotic touch occurred in October when a team of engineers from Stanford University’s Zhenan Bao Research Group succeeded in creating an artificial nerve. The man-made device was designed to operate just like its human counterpart in that it would allow robots to respond to touch effectively.

The system relies on an artificial nerve setup that can be broken into three components. The mechanoreceptors act as resistive pressure sensors. There are also organic ring oscillators, which function similarly to neurons, and organic electrochemical transistors that enable the entire system to operate.

로봇 스웨터

Carnegie Mellon University introduced a Robot Sweater that could make industrial robots much safer. Notably, today’s safety systems often require rigid pieces added to robots. The problem with his approach is that robots can’t be fully covered as their moving parts need to remain flexible. This desire led researchers to consider a sweater-like fabric as the solution.

The Robot Sweater is a machine-knitted cover that can fit any 3-dimensional shape. As such it can be created to provide full protection for robots and their human co-workers. The device works by utilizing two layers of metallic fibers integrated into its surface. Whenever a human touches the sweater, it closes the circuit, notifying the robot of the incident and inciting a response.

촉감 로봇의 미래

Robots that can emulate touch are the future. These devices will open the door for more integration alongside human coworkers. This development will lead to robots that improve safety,  save funding, and provide multiple industries with solutions to long-lasting problems. As such, demand for robots that can emulate touch is only going to increase over the coming years. For now, you should expect to see more robot coworkers in the coming months.

Learn about other cool robotics projects now.

연구 참고:

^{1. Tafone, D., McEvoy, L., Sua, Y. M., & Huang, Y.-P. (2024). Surface roughness metrology with a raster scanning single photon LiDAR. Applied Optics, 63(30), 7917–7923. https://doi.org/10.1364/AO.537404}