Employing 3D virtual reality games to develop ANN for device control: a pilot study

Non-immersive virtual reality (VR) game scenarios were developed to aid in the collection of EMG parameters from the biceps and triceps while subjects performed a sequenced series of tasks in the virtual environment. For each subject the best ANN configuration (combination of hidden layers and transfer functions) was chosen, with the resulting optimized algorithms used to classify the sequence of contractions and the function type of the subjects while playing new game scenarios. The wide variety of individually configured ANN developed show why it is difficult to train new users of myoelectric devices with a single algorithm. The use of VR-based games shows promise as a training technique for individuals needing to develop control for prosthetic limbs.