STEERING CONTROL ASSISTANCE FOR A POWERED WHEELCHAIR USING DEEP NEURAL FRAMEWORKS

Abstract

People with intellectual, motor, sensory damage always faces difficulty in their safe mobility. Thus, for their safe and secure navigation caregivers are needed always. To help these disabled persons for an independent mobility, a new deep learning technique for steering control assist for a powered wheelchair is proposed. In this work, the system learns to direct a powered wheelchair with user satisfaction in new environments. Long-Short Term Memory (LSTM), Bidirectional LSTM (BLSTM ) and Gated Recurrent Unit (GRU) neural networks are used for assisting the guidance of a powered wheelchair. The inputs to these neural networks is dataset of ranges produced by ultrasonic sensors considered to be devoted on the wheelchair and output is the six direction classes. The model predicts the wheelchair direction on the basis of the input data. Later the direction predicted by neural network techniques is been combined with user defining direction which produces the resultant direction so as to avoid objects in the path. The differently-abled person or patients, uses an input method to provide desired speed along with direction and the neural network provides a safe direction for the wheelchair avoiding objects in vicinity. A powered wheelchair model simulation is also done and results are verified with the expected outcomes. The comparison study on numerous parameters concludes that GRU model showed the best performance with overall accuracy of 97.38%. The accuracies obtained from LSTM and BLSTM models are 95.17% and 95.89% respectively