Development and Validation of a Low-Cost Optical Sensor System for Real-Time Teleoperation of Anthropomorphic Robotic Hands
Abstract: Dexterous robotic hand teleoperation demands data gloves balancing sub-degree measurement accuracy with sub-30-millisecond response latency—capabilities historically limited to commercial systems costing $5,000 to $30,000. These prohibitive costs restrict access for research laboratories, educational institutions, and small scale developers, constraining innovation in surgical robotics, hazardous environment teleoperation, and rehabilitation applications.
This thesis demonstrates that research-grade performance can be achieved at consumer-grade cost through infrared optical bend sensing. The developed sensor employs an 850 nm LED and photodiode embedded in silicone tubing, where light attenuation varies exponentially with joint curvature. A transimpedance amplifier with 100 kΩ feedback resistance converts photocurrent to voltage, sampled by a Teensy 4.0 microcontroller at 1 kHz. Logarithmic calibration mapping θ = k · ln(Vref/Vout) relates voltage to joint angle.
Comprehensive validation against three independent measurement systems—optical sensor, LEAP Hand Dynamixel encoders, and Arducam 4K camera vision tracking—demonstrates strong correlation (Pearson r = 0.9957) across the 0–45° operating range. Extended testing over 325 seconds (1,931 synchronized samples) reveals measurement error of −0.024 mean with 1.015° standard deviation and 1.08° RMSE. Additional characterization quantifies hysteresis (0.96° average width), progressive drift (−0.63 over 1,000 cycles), and temperature sensitivity, establishing performance boundaries for practical deployment.
The complete system costs under $50 per sensor, enabling full 15-sensor hand coverage for approximately $750 compared to $5,000–$30,000 commercial alternatives—ai 7-to-40-fold cost reduction while maintaining comparable accuracy. This democratization of precision hand tracking technology removes economic barriers to research in surgical robotics, teleoperation of manipulators in hazardous environments, and quantitative rehabilitation therapy. Comprehensive documentation and open-source hardware design facilitate reproducibility and community-driven refinement, expanding access to high-fidelity human-robot interaction capabilities previously limited to well-funded laboratories.
Advisor: Professor Cagdas Onal
Committee: Professor Jane Li (Robotics Engineering), Professor Fiona Yuan (Robotics Engineering)