Fall Prediction Model for a Reconfigurable Mobile Support Robot

Abstract

This work presents a fall prediction model to be used in conjunction with a reconfigurable robot for elderly mobility support. The fall prediction model is based on a Long Short Term Memory network. A predicted fall will inform a reconfigurable robot to expand its base of support to avoid possible tipping induced by the fall. A wearable support interface consisting of an instrumented harness and auto retracting cable system is developed for supporting the body and preventing a fall from occurring. The prediction model was developed using data taken of simulated falls and activities of daily living while using a test platform with the wearable support interface solution. The reconfigurable robot concept explored resembles a walker and provides mobility assistance during normal use, but it can also expand its base of support during a falling emergency. The model results show that a fall can be predicted 530 ms from the initial observance of instability in the user, which allows sufficient time to reconfigure a robot into a more stable configuration.