Meta is betting on developing embodied AI, announcing new steps to integrate the sense of touch into its robotic innovations. Meta has partnered with U.S.-based GelSight and South Korea’s Wonik Robotics to develop and market tactile sensors for AI, targeting innovations toward scientists and researchers rather than individual consumers, according to Interesting Engineering.
Recently, Meta launched three major research components – Sparsh, Digit 360, and Digit Plexus – focusing on tactile perception to enhance robot skills and human-robot interactions. Additionally, it introduced a new standard called “PARTNR” to evaluate planning and logic in human-robot collaboration, contributing to making robots skilled partners.
Sparsh, developed in collaboration with the University of Washington and Carnegie Mellon University, is a set of AI models that grant robots the ability to perceive touch and determine appropriate pressure on objects. Digit 360 is a tactile sensor shaped like an artificial finger, equipped with over 8 million “taxels,” allowing it to capture multi-directional shapes precisely, and it enables rapid touch response thanks to localized data processing.
Digit Plexus is a hardware and software platform that allows the integration of multiple tactile sensors for robotic fingertips, efficiently transmitting touch data to a host computer via a single cable. Meta aims to assist researchers in advancing robot dexterity faster by publishing Digit Plexus codes and designs.
Meta also launched the PARTNR standard, a new benchmark to assess the ability of AI models to assist with household tasks and achieve effective human-robot interaction. Meta stated that the goal of PARTNR is to transition AI models from being “agents” to “partners,” driving progress in collaborative intelligence and facilitating advanced human-robot interaction.
