Robotic vision degrades exactly where precision matters most: poor focus at millimetre scales, occlusion, glare and low light all undermine cameras in settings such as space hardware, radioactive zones and hazardous industrial sites. Research from Yonsei University and the University of Southern California, published in Microsystems and Nanoengineering in March 2026, proposes a division of labour, letting vision find the object and touch finish the job.
In the system, a single RGB-depth camera locates an object and estimates its position, size and overall shape, performing strongly in good light, with detection accuracy around 0.995 above 100 lux, but dropping to about 0.706 below 10 lux. A soft, inkjet-printed 10 by 10 pressure sensor array then scans the surface directly, reconstructing fine features and three-dimensional profiles that the camera cannot resolve. The tactile array registers loads as light as roughly 23 pascals, responds in about 48 milliseconds, and maintained performance over 5,000 loading cycles.
In tests the combined approach distinguished height differences of 2, 3 and 4 millimetres on irregular surfaces, the kind of discrimination needed to inspect or manipulate small components reliably. The broader message for robotics researchers is architectural: rather than pushing ever more capability into vision alone, pairing modest cameras with high-resolution printed touch offers a cheaper, more robust route to fine perception in environments where light cannot be trusted.