An Intention Prediction Based Shared Control System for Point-to-Point Navigation of a Robotic Wheelchair

Abstract

Shared control approaches for robotic wheelchairs aim to provide navigation assistance to humans by utilizing robot’s intelligence in environment perception and motion planning. They can be broadly classified into two categories based on human intention prediction. Without human intention prediction, control authority lies with humans and assistance is provided only to avoid collisions. This can cause difficulty in cases where fine motor control is required, such as when entering narrow doorways, especially for users with severe upper limb disability. Intention prediction based approaches are able to better assist with such tasks but do not give enough control authority to the user as possible user intentions are pre-defined. In this work, we present an intention prediction based shared control system for point-to-point navigation of wheelchair which gives control authority to the user and also assists in fine motor control tasks. We compute various possible user intentions online using generalized Voronoi diagram, link them across time steps using their <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">homotopy class</i> and thus are able to calculate their probability given user input history. A shared local path planner then steers user towards the most likely path. This allows the user to follow any path. Our simulation experiments with 18 healthy subjects and both simulation and real wheelchair experiments, with 2 Cerebral Palsy (CP) subjects, show that our system can improve navigation outcome for people with disability and in general leads to around 10% faster completion of the task for even healthy people as compared to a local obstacle avoidance system while allowing users to follow their desired path with similar accuracy.