Soft optical sensor enables multi-directional posture sensing in robotic hands

Apr 01, 2026

Researchers developed a soft optical bending sensor that enables humanoid robotic hands to perceive finger posture across multiple axes during dexterous tasks.

(Nanowerk News) A team of engineers has developed a humanoid robotic hand equipped with a new type of soft optical sensor that can track finger posture across multiple directions simultaneously. The soft bending sensor, embedded into a dexterous hand with 18 active degrees of freedom, allows real-time perception of both flexion and lateral finger movement during delicate manipulation tasks. The work, published in Microsystems & Nanoengineering (“Soft sensor for omnidirectional posture perception in humanoid dexterous hands”), addresses a persistent limitation in robotic manipulation: the inability of existing soft sensors to decouple different bending motions when fingers flex and move sideways at the same time.

Key Findings

Robotic hands have become increasingly capable at grasping and pinching objects, but most still lack the subtle motor control that makes human hands so adaptable. A central challenge is proprioception, the body’s internal sense of limb position and movement. Human fingers continuously monitor their own posture, allowing fluid adjustments during complex tasks. Most robotic hands, by contrast, rely on sensors that detect bending in only one direction or produce coupled signals when a finger moves along two axes at once, making true dexterous manipulation difficult to achieve. Researchers from Zhejiang University, Hangzhou Dianzi University, and Lishui University set out to solve this problem with a sensor design that separates directional signals by optical means. Their humanoid hand features five rigid-flexible fingers, each fitted with a soft sensor constructed from segmented PMMA (polymethyl methacrylate) optical fibers, a trichromatic LED, and a chromatic detector. The overall system provides 18 active degrees of freedom, matching the kinematic complexity needed for tasks that go beyond simple gripping. Systematic overview of the humanoid dexterous hand with multi-degree-of-freedom posture perception. a The humanoid dexterous hand in various delicate operational scenarios. b The omnidirectional soft bending sensor in the humanoid dexterous hand. (Image: Reproduced from DOI:10.1038/s41378-026-01179-3, CC BY) The operating principle relies on differential light attenuation. As the sensor bends, red, green, and blue light traveling through the segmented fibers lose intensity at different rates depending on the direction of bending. Because the fiber geometry separates the optical responses associated with pitch (flexion) and yaw (lateral movement), the system can distinguish the two motions rather than conflating them into a single mixed signal. This decoupling is what sets the design apart from earlier soft sensors that struggled with multi-axis interference. Bench testing confirmed strong repeatability across 100 bending cycles, with root mean square error values of 2.1%, 1.9%, and 3.2% for the three optical channels. Under single-axis bending, the sensor achieved average measurement errors of ±2.13° for pitch and ±2.34° for yaw. Crosstalk between axes was low: pure yaw motion contributed only 3.2% to the pitch reading, while pure pitch contributed 4.1% to the yaw reading. Signal-to-crosstalk ratios reached 50.68 dB and 30.81 dB for the respective axes, indicating that each channel captures its target motion with minimal interference from the other. Beyond laboratory characterization, the team demonstrated the hand in three tasks designed to test fine motor coordination rather than raw grip strength. The hand cut paper with scissors, clicked a computer mouse, and pressed individual piano keys, all under closed-loop posture control. Each task required precise, coordinated adjustments across multiple finger joints, confirming that the sensor provided sufficient real-time feedback to guide subtle movements during operations that would defeat a conventional robotic gripper. The researchers note that the integrated rigid-soft mechanical design supports natural finger movement, while the optical sensing system delivers the stability, repeatability, and multi-degree-of-freedom posture perception required for complex operations. It is this combination of mechanical compliance and accurate multi-axis sensing, rather than either element alone, that constitutes the core contribution of the work. Improved posture perception in robotic hands could have practical implications in several areas that the researchers identify explicitly. Humanoid robots used in service environments, industrial assembly lines, and rehabilitation devices all require fingers that can adapt to fragile or irregularly shaped objects. The demonstrations also point to applications in human-robot interaction, where smoother and safer hand motion is critical. By showing that soft optical sensing can maintain accuracy during complex, multi-directional finger movements, this work brings robotic manipulation measurably closer to the adaptive capability of the human hand.