![AI editorial illustration: two-finger robotic gripper gently holds a raw egg in a lab [Image content created with AI]](https://alpha-bionic.info/wp-content/uploads/2026/10/tactile-gripper-egg-editorial-1.png "Daimon Robotics: Tactile Sensors, Data Gloves and AI for Dexterous Robot Hands 1 [Image content created with AI]")
The visit to Daimon Robotics in Shenzhen is about far more than an unbroken egg. In Everlast Robotics’ video, tactile sensors, robot fingertips, a data glove and the shortage of useful training data all matter. Demonstrations involving a Chinese character, a cable, a key and a pen placed in a case illustrate different manipulation problems. None is equivalent to proving that an entire humanoid can handle any unfamiliar object autonomously.
In this article
Daimon Robotics: a hardware–data–model strategy
The Shenzhen company says it grew out of research at the Hong Kong University of Science and Technology. It develops vision-based tactile sensors, grippers and tools for collecting manipulation data. Its company profile and technology overview describe a chain from sensor hardware to physical-interaction datasets and tactile AI models. The essential question is how measurements at the point of contact become repeatable robot skills.
How a fingertip sensor can “see” force
An external camera can locate an object before contact, but the fingers often hide the contact patch once grasping begins. In a vision-based tactile sensor, a miniature camera inside the sensor observes deformation of a compliant surface. Software can estimate contact geometry, deformation and force-related signals. A controller can then change grip, ease pressure or detect an emerging slip. This is feedback for a control loop, not dexterity created by pixel count alone; a sensing point is not necessarily a separate physical force transducer.
The video includes a character-recognition demonstration and later shows a cable, key and raw egg. These require different capabilities: resolving fine surface features, controlling contact during motion, and applying enough force to hold without damaging.
40,000 points per square centimetre versus 110,592 total points
One video chapter cites 40,000 sensing units per cm². Daimon used that area-density figure for the earlier DM-Tac W in its product announcement. Its current DM-Tac W2 specifications instead list 110,592 effective points in total and 120 Hz sampling. The W2L and W2M have different sizes, weights and sensing areas. Points per unit area and points across a complete sensor are not interchangeable measurements.
The smaller DM-Tac F fingertip is specified at 180 Hz and senses across the pad, side and tip. The video does not make it possible to identify with certainty the exact sensor variant in every shot. A claim such as “160 times better than a human finger” also depends on which resolution or sensitivity is being compared. It does not mean 160 times the versatility of a human hand.
Why cables, keys, eggs and a pen matter
A cable must be guided without kinking or slipping. A key demands the right contact points and orientation. An egg needs a narrow force window: too little pressure and it drops, too much and it breaks. The pen-in-a-case demonstration adds several stages—pick up, align and insert. Transitions from free movement to contact are particularly difficult for robots.
The footage does not provide a full trial record: failures, object variation, cycle times and the degree of human control are unclear. Nor does the video’s question about the price of a “feeling hand” establish a generally valid purchase price. A sensor, gripper, dexterous hand, controller, integration and maintenance are separate cost items.
The data glove, teleoperation and the simulation gap
Another chapter examines a data glove. During teleoperation, a person performs a task while the system records movement and, ideally, contact and force information. Those demonstrations can provide training examples for robot control. Daimon has announced the DM-EXton2 teleoperation data-collection system with haptic feedback.
More recordings alone are not sufficient. Useful datasets need varied objects, starting positions and failures, synchronized sensor streams and evidence that a learned skill transfers to other hands. Simulation can generate many variants cheaply, but real friction, flexible materials and contact transitions are hard to model faithfully. The video’s data discussion points to this gap between a demonstration and scalable autonomy. Daimon positions its data infrastructure and tactile world model as its answer; that remains a company technology claim, not independent proof of general-purpose manipulation.
Customers, hospitals and what an industrial buyer should ask
The video description claims more than 200 customers and names OpenAI, Google DeepMind and BMW. It does not publicly document the count or the nature of each relationship. A sensor shipment, a research evaluation and a productive factory deployment are very different forms of adoption. Likewise, a possible hospital application should not be confused with a clinically approved system.
For manufacturers, the question is narrower: which part is currently damaged or frequently missed? Can tactile feedback lower scrap rates at the required cycle time? How often do operators intervene? How well does the sensor survive wear, cleaning and recalibration? What happens after a tool change? A dedicated two-finger gripper may be more practical for such a task than an entire humanoid. Broadly human-like manipulation remains a different benchmark, as roboticist Rodney Brooks has argued.
Frequently asked questions
Can humanoid robots now reliably grasp raw eggs?
The video shows a specific tactile grasping demonstration. Repeatable, autonomous performance across unfamiliar objects is not established.
Do more sensing points mean a human-like sense of touch?
No. Sensor resolution is only one factor. Mechanics, calibration, control-loop speed and transfer to new tasks also matter.
What evidence is still missing?
Documented repeated trials with success and breakage rates, varied objects and placements, a clear account of autonomy, plus speed, durability and cost data.
Image note: The featured image is an AI-generated editorial illustration of a mechanically plausible two-finger gripper, not a photograph of a Daimon product or the video demonstration.
![Nico Nuss [Image content created with AI]](https://alpha-bionic.info/wp-content/uploads/2025/12/Nico-Nuss_1-150x150.jpg "Daimon Robotics: Tactile Sensors, Data Gloves and AI for Dexterous Robot Hands 2 [Image content created with AI]")
Author Nico Nuss has worked with mobile computing and automation software since 2001. His many years of experience and strong interest in emerging technologies shape his work on robotics and artificial intelligence.