Estimating powered wheelchair driver intentions more accurately using force feedback information

Abstract:
Nowadays society’s interest in and need for robots that assist people in executing difficult tasks such as navigation and manipulation is only increasing. In earlier work, we have developed a Bayesian framework to provide navigation assistance to powered wheelchair drivers. Our framework consists of two major parts, a part that estimates the driver’s unknown navigation intention and a part that modifies the driver’s steering signals or gives user feedback based on the estimated intention in order to obtain safe and easy driving. A key component in both parts is a probabilistic user model, which can capture complex user-dependent driving behaviour. This work focuses on extending our user model, previously adopted for drivers that use a proportional joystick, to interfaces that provide interaction force information as well. We show that in our experiments, force information alone yields less accurate information about the driver’s navigation plans as compared to the proportional joystick’s velocity information. However, the combination of force and velocity information increases the accuracy of our intent estimation component.