Appearance-based Navigation Capabilities for a Semi-Autonomous Control of Powered Wheelchairs

点击翻译按钮获取中文摘要

Cyberworks Robotics inc. is developing an autonomous navigation systems for power wheelchairs. Its system uses a RGB-D sensor (a Kinect V2 camera) combined with low cost ultrasonic range sensors, laser ranger finder, and odometry (using wheel encoders an Inertial Measurement Unit (IMU)). Its development is based on RTAB-Map for Simultaneous Localization and Mapping (SLAM). Developed at IntRoLab of the Université de Sherbrooke (UdeS) and distributed open source, RTAB-Map handles online appearance-based loop closure detection, i.e., the ability to visually detect that the robot is visiting an area that it has seen before, for large-scale and long-term operation. Cyberworks Robotics approached IntRoLab to examine hot to make its current implementation more robust by using visual inertial odometry and other optimizations for real world conditions such as feature sparse, feature repetitive and dynamic conditions. To do so, IntRoLab plans to adapt its recent development, SPLAM-MM (Simultaneous Planning, Localization And Mapping), which allows a robot to continuously update its map to incorporate dynamic changes of the environment. The objective consists of extending and validating its use to exploit visual inertial odometry and to deal with situations like loss of visual tracking during turns, moving people in the scene, false loop closures, feature repetitive environments, feature sparse environments, adapting to minor and major real time changes in the environment, shared human control, trajectory planning for traversal of narrow doorways, dynamic obstacle avoidance path re-routing, and address potential failure conditions such as intense IR sources like intense direct sunlight, heating radiators, and around glass/chrome stairs/railings, etc. The underlying purpose of such self-navigating system is to assist powered wheelchair users such as quadriplegic, amyotrophic lateral sclerosis and multiple sclerosis, who have difficulty using traditional power chair joysticks or alternative input devices.