Knee & femur joint angle estimation based on mechanical & electromyographic measurements for active control of transfemoral prostheses

The advancement of robotics has given way to active prosthetics for aiding those with limb amputations. Voluntary control of the said prostheses requires the prediction of the corresponding healthy limb joint. Therefore, this paper has attempted to utilize Artificial Neural Networks (ANNs) to analyze numerical features extracted from acceleration, angular velocity, force and electromyographic signals to accomplish this through supervised learning. The true values of the target variables were recorded by a regular phone camera at 30 frames per second, final r$^{2}$ score of 89 percent was achieved for both knee and femur joints with mean absolute error of 0.16 and 0.13 radians for knee and femur joints, respectively.