Researchers at New York University have developed an affordable, durable sensor that’s easy to assemble and integrate into robots.
Creating cost-effective, high-performance sensors holds significant potential for robotics research by enhancing robot perception, which could improve manipulation and navigation capabilities.
Researchers at New York University have developed AnySkin, an affordable and durable sensor that is simple to assemble and integrate into robotic systems. The sensor is more accessible than many other tactile sensors recently introduced, potentially broadening the scope of robotics research.
AnySkin, a new magnetic tactile sensor developed by the team, is an enhanced version of an earlier sensor they introduced, known as ReSkin. This updated sensor maintains the straightforward design of ReSkin but improves upon it with more consistent signal output and a distinct separation between the electronic components and the sensing interface.
AnySkin is designed for quick assembly, taking only a few seconds, and can be used to train artificial neural network models with minimal to no preprocessing required. It outperforms ReSkin in collecting tactile signals more consistently and offers the advantage of easy and rapid repairs in case of accidental damage.
The AnySkin sensor features a unique self-adhering design that allows for flexible integration. It can be easily stretched and affixed to different surfaces to provide them with sensing capabilities.
This sensor is also remarkably versatile; it can be fabricated in various shapes and sizes and assembled with ease. Additionally, AnySkin can be effortlessly peeled off and replaced if damaged.
In preliminary testing, researchers discovered that their sensor, AnySkin, performed impressively, achieving results on par with other established tactile sensors. They also noted that various AnySkin sensors demonstrated consistent performance and sensing responses, indicating they can be reliably reproduced and utilized on a large scale.
Looking ahead, AnySkin could be applied across a broader array of robotic systems and evaluated in more diverse scenarios. The researchers believe it is ideally suited for gathering extensive tactile data, which could be used to train large-scale deep learning models akin to those used in computer vision and natural language processing (NLP).
Resources: Raunaq Bhirangi et al, AnySkin: Plug-and-play Skin Sensing for Robotic Touch, arXiv (2024). DOI: 10.48550/arxiv.2409.08276
