DESIGN AND DEVELOPMENT OF A LOW-COST WIRELESS DATA GLOVE FOR REMOTE ROBOTIC HAND OPERATION

摘要

The goal of this project is to develop a teleoperated robotic hand with an emphasis on minimizing cost. Traditionally, use of low-cost sensors meant a sacrifice in the positional accuracy of the finger joints due to noise and high degree of nonlinearity. External vision systems provide accurate data, but suffer from being limited to specific lighting conditions and are hardly mobile. We have developed a robust data glove with a wireless connectivity to a nearby PC that is networked to a robotic arm in the different laboratory through a wireless LAN. In its prototype stage, each finger is modeled as a single joint. To measure the relative angular displacement of each joint, a hollow transparent polyurethane tube is attached along the length of each finger. An Infrared LED and a matched phototransistor are attached at the opposite ends of each tube. The bending of each finger proportionally diminishes the intensity of the light seen by the phototransistor, as a fraction of emitted energy escapes through the bend of the transparent tube. To filter the affects of external lighting and mains coupling each IR source is modulated at different frequencies to prevent interference between the signals. The relationship between relative angular displacement and the sensor signal is a highly nonlinear function that we have linearized through a combination of a non-linear amplifier and a digital look-up-table (LUT). The LUT is generated by custom-designed computer vision software that fits the true angular displacements to the function generated by passing the signal through a non-linear amplifier stage. A low-cost onboard 8-bit microcontroller applies the stored LUT to the digitized measurements from the sensors. The 16-bit positional values are then transmitted by a UDP (Universal Data Protocol) to the client PC that outputs motor commands to the servos in the robotic hand. Positional accuracy, noise and linearity were then analyzed by the same machine vision software and its effectiveness compared to the pure machine vision approach.