Rubbery, Multimodal-Sensorized and Neuromorphic-Computing E-Skin for Intelligent Soft Robots
Abstract
The development of intelligent soft/humanoid robots that can team with humans and perform dexterous tasks largely centers around the realization of two capabilities: 1) highly accurate, high-resolution and multi-modal sensation to the interactions with surrounding environments; 2) human-like intelligence, i.e., artificial intelligence (AI), for high-level information extraction from the sensing data and then decision-making in performing complicated tasks. For realizing these essential capabilities on soft-bodied robots that can provide safe human interactions, all the hardware devices for providing the sensing and AI-computing functions need to be soft, stretchable, and providing unaltered performance under actuated motions on robots. For achieving human-mimetic tactile esture sensing, there have been some efforts in developing stretchable pressure and strain sensors. However, there are still two major gaps towards realizing the desired sensing accuracy and resolution on humanoids robots: 1) the lack of non-interfered tactile sensing under skin deformations (i.e., stretching and bending) from motions; 2) the lack of high-resolution, multimodal integrations. On the other hand, all the demonstrations in applying machine-learning algorithms for analyzing sensory data from electronic-skins on robots have been carried out through transferring the data to centralized computing units, which has several major limitations. Alternatively, to realize the efficient implementation of machine-learning data analysis locally on robotic e-skins, it is highly desirable to endow skin-like tactile sensory networks with near-sensor or in-sensor machine-learning computing. In this proposal, Dr. Sihin (namely, NeuroSense Skin) for intelligent soft robots, so as to enable automated learning of complicated interactions with environment. so as to enable automated learning of complicated interactions with environments. Specifically, through exploiting and combining the advances in stretchable polymer electronics and neuromorphic computing, we propose to develop a suite of innovative device concepts, designs and fabrication approaches for realizing unprecedented but highly desired sensing and machine-learning-computing capabilities: 1) a stretchable, strain-unperturbed and bending-unperturbed pressure sensor that can provide motion-interference-free tactile sensing; 2) a high-density array of multi-modal tactile and body-motion sensors for pressures, strains, and incipient slips;3) a stretchable neuromorphic transistor array for machine-learning-based near-sensor computing; 4) a new in-sensor computing device concept of a non-volatilely, reconfigurable pressure-sensor array that has a built-in artificial neural network (ANN) for simultaneous sensing and data processing. In the end, our NeuroSense Skin will be integrated with a pneumatically-actuated soft robotic hand capabilities that are highly relevant to Navys missions. Our NeuroSense Skin will significantly enhance soft robots perception and intelligence for recognition, training and learning, which could greatly accelerate the expansion of robots functions in naval tasks, including human-robot teaming, independent task execution in remote environments, and cognitive control by autonomous systems.Its applications on human bodies or biological organisms will also generate technological merits for both injured and healthy warfighters, and for the study of bionic aquatic swimming. In addition, the innovative AI computing platform can greatly elevate the practical applicability of AI to massive internet of things (IoTs) systems for transforming vast data into time knowledge.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jun 09, 2021
- Source ID
- N000142112581
Entities
People
- Sihong Wang
Organizations
- Office of Naval Research
- United States Navy
- University of Chicago