A Robot Hand That Feels to a Tenth of a Newton

Synaptics has brought its Capacitive Tactile Sensing module and Astra edge AI processors into NVIDIA's Isaac Sim and Holoscan workflows, announced 17 September.

The sensor specification is concrete. A 5 by 12 taxel array at 2.5 mm pitch, a force range up to 100 N with 0.1 N resolution, and a rebound response stated as ten times faster than traditional foam sensors. The SN6012T touch controller drives the module, and it now streams to Holoscan over standard Ethernet — which means the sensor can sit on the tool while the compute node sits elsewhere on the plant network rather than on the arm.

Rahul Patel, Synaptics CEO, framed the combination as processing multi-sensory modalities and simplifying integration hurdles. Target applications are dexterous robotic hands, grippers, physical AI systems, humanoids and industrial automation needing precise grip control.

Two numbers explain why this is more than a demo. A resolution of 0.1 N is roughly the difference between holding a moulded plastic part and marking it, which is the practical threshold for handling finished goods that a customer will inspect. And Isaac Sim support means the grip policy can be developed and trained against a simulated sensor before the hardware exists, which changes the project sequence — tactile behaviour has historically been the part that could only be tuned on the physical cell, late, with the line waiting.

Vision solves where an object is. It has never solved how hard to hold it, and lines that handle soft, deformable or fragile goods have mostly worked around that with fixtures rather than feedback.