Compact Optical Single-axis Joint Torque Sensor Using Redundant Photo-Reflectors and Quadratic-Programming Calibration

A team from KAIST (Hyun-Bin Kim, Byeong-Il Ham, Kyung-Soo Kim) has developed a breakthrough optical torque sensor that replaces problematic current-based sensing in robot joints. Current sensors in collaborative robots struggle with poor accuracy at low torque levels due to gearbox stiction and nonlinear current-torque relationships. The new design uses a non-contact approach, measuring micro-deformations in an elastic structure with a redundant array of photo-reflectors arranged in four directions. This optical method dramatically improves sensitivity and signal-to-noise ratio.

The sensor's compact form factor (96mm diameter, 12mm thickness) makes it practical for real-world robotics applications. The key innovation is a quadratic-programming-based calibration method that exploits sensor redundancy to suppress noise, achieving a 3 sigma resolution of 0.0224 Nm at 1kHz without filtering—a 2.14x improvement over traditional least-squares calibration. Experiments show maximum error of just 0.083%FS and RMS error of 0.0266 Nm for z-axis torque measurement.

Temperature compensation through rational fitting addresses drift issues from MCU self-heating and motor heat, ensuring stable performance. Motor-level validation demonstrates superior low-torque tracking and disturbance robustness compared to current-sensor-based control in both torque control and admittance control scenarios. This technology enables safer physical human-robot interaction and more precise manipulation tasks.