Sharpa Equips Robots with 'Full-Body Touch': What's the Real Strategy Behind the New Hardware?
Sharpa is building a data flywheel using data gloves and dexterous robotic hands, translating the manual skills of veteran workers into tactile feedback for robots. The deployment logic of embodied AI is shifting from purely vision-based positioning to full-body somatosensation.
At the IROS conference in late September 2026, Sharpa launched three flagship products: the D01 robot, the W02 dexterous hand, and the AE01 data glove. Shortly after, on October 7, the Economic Development Board (EDB) of Singapore published an interview with its founder. The company's ambition is clear: to evolve robots from "grasping by sight" to "grasping by touch".
What Changes When You Give a Robotic Arm a Nervous System?
Sharpa's core focus is reportedly "dexterous manipulation." While robots have traditionally relied on vision-based positioning to perform tasks, they are now evolving to possess "full-body somatosensation" (the sense of touch and body position). Simply put, this means robots will no longer just track where a part is; they will be able to sense the pressure, slip, and even the texture of the part they are holding. In a way, this is like giving a cold mechanical arm a nervous system. For the workforce, this represents more than just a production line upgrade. It means that robots can finally take over highly delicate tasks that currently require extreme manual dexterity and rely on the "intuition" of veteran technicians.
Digitizing a Master's Skills in Three Steps
Together, these three products form a complete "data flywheel," with deployment occurring in three steps: First, a veteran human worker wears the AE01 data glove to perform physical tasks. Second, the system records finger micro-adjustments and applied force in real time. Third, the W02 dexterous hand and D01 robot directly mimic and execute the movements. Imagine a scenario in an automotive engine plant: a veteran fitter, Master Li, wears a data glove to manipulate irregularly shaped parts, and every micro-adjustment of his fingers is recorded. It is like assigning a personal tutor to the robot, digitizing the worker's "muscle memory" for the robot to imitate. However, there is a catch: if Master Li teaches the robot to grip a metal part, the robot might fail completely if switched to a smooth plastic part or a different rubber material. "Muscle memory" trained on specific irregular parts cannot necessarily be transferred to other objects. It is also worth noting the physical wear and human friction involved in this "learning" process. Prolonged, high-frequency use of data gloves can lead to sensor degradation, causing the collected data to become inaccurate. More practically, asking veteran employees who are used to working bare-handed to wear gloves daily for data collection involves cooperation willingness and extra training costs—a calculation that small and medium-sized manufacturing enterprises must consider when adopting such solutions. If cross-scenario generalization fails, these dexterous hands will be confined to acting as "skilled workers" on a single production line.
Choosing Singapore: It's Not About Cheap Labor
In the EDB interview, Sharpa's founder revealed the reasons for choosing Singapore as the launchpad for its embodied AI. Many assume the choice was driven by manufacturing capabilities, but that is not the case. Rather, Singapore provides a "super testing ground." When deploying embodied AI in the real world, the biggest fear is a lack of liability coverage in case of accidents. Singapore boasts highly clear regulations and a relatively tolerant social attitude toward new technologies. Unlike traditional manufacturing companies going global solely for cheap labor and land costs, the site-selection logic for AI companies has changed. This is similar to Chinese new energy vehicle (NEV) manufacturers building factories in Europe—they are looking not just for market access, but for mature local regulatory frameworks. In the future, the competition in embodied AI will rely not only on supply chain depth but also on an institutional environment that "allows for mistakes" and clear data compliance. By using Singapore as a primary base, companies can iterate quickly in a well-regulated environment rather than trial-and-error in ambiguous gray areas.
From Rote Execution to Generalization: How Far Are We from a Universal Robot Butler?
The entire industry is undergoing a paradigm shift: moving from "pre-programmed execution" to "experience imitation." Previously, robots followed rigid code; now, they learn from human experience. It is like shifting from "cooking strictly by a recipe" to "tasting and adjusting the seasoning." However, cross-scenario generalization remains the core variable for the future. This means we are still some distance away from a "universal robot butler," but in specific industrial scenarios, this "experience imitation" can already tangibly replace some high-intensity, repetitive labor. For manufacturing managers, the practical approach is not to expect a single robot to handle all processes, but to first deploy it on a single workstation that requires the most manual dexterity and get it working perfectly there.
Key Takeaway: In the deployment of embodied AI, hardware is just the entry ticket. The real capability lies in seamlessly converting the "muscle memory" of veteran workers into a data flywheel, and finding a testing ground with a high tolerance for errors.