RESUMO
This paper describes a novel Deep Learning architecture to assist with steering a powered wheelchair. A rule-based approach is utilized to train and test a Long Short Term Memory (LSTM) Neural Network. It is the first time a LSTM has been used for steering a powered wheelchair. A disabled driver uses a joystick to provide desired speed and direction, and the Neural Network provides a safe direction for the wheelchair. Results from the Neural Network are mixed with desired speed and direction to avoid obstacles. Inputs originate from a joystick and from three ultrasonic transducers attached to the chair. The resultant course is a blend of desired directions and directions that steer the chair to avoid collision. A rule-based approach is used to create a training and test set for the Neural Network system and applies deep learning to predict a safe route for a wheelchair. The user can over-ride the new system if necessary.
Assuntos
Aprendizado Profundo , Pessoas com Deficiência , Cadeiras de Rodas , Desenho de Equipamento , Humanos , TransdutoresRESUMO
Research is described that determines the direction that a powered wheelchair will take, using the preference ranking organization method for enrichment of evaluations II (PROMETHEE II). This is the first time that this sort of decision making has been used in this type of application. A user suggests a desired direction and speed, and the decision-making system suggests a safe direction for a wheelchair. The two are mixed, and the wheelchair then tends to avoid obstacles. The inputs come from a powered wheelchair joystick and two ultrasonic transducers fixed onto the wheelchair, and the final direction is a simulation of the desired direction and a direction that moves the wheelchair away from obstacles. The arrangement assists a disabled driver to steer their wheelchairs. It uses a systematic decision-making process, and this paper presents the process and a new way of selecting a best compromise direction. Sensitivity analysis is employed to explore potentially suitable directions as uncertainty and risk may be present. A suitable direction is selected that provides a robust solution. The user would be able to override the suggestions by the PROMETHEE II system by holding the joystick in a position.