RGB - based Obstacle Detection and Avoidance for an EMG - Controlled Wheelchair

Kurzfassung

Many people suffer from limb functionality impairment due to neuromuscular disease, stroke or trauma. This strong reduction of independency makes 24-hour care necessary. Various assistive devices have been developed which provide support in every-day tasks and enhance mobility. Focusing on wheelchairs, many different steering methods are provided including joysticks and human-machine interfaces (HMI). However, controlling a mobile platform through an HMI can be highly difficult and inaccurate. Shared control strategies support steering through HMIs with assistance during difficult maneuvers. This work enhances the shared control approach of the mobile wheelchair 'EMG-controlled Daily Assistant' (EDAN). An obstacle detection and avoidance facilitates the steering to the user. The Obstacle Detection segments objects from a depth image provided by an RGB-D camera and creates a 2D obstacle map. The Obstacle Avoidance adjusts the steering commands of the user to circumnavigate obstacles. Based on an Artificial Potential Field repulsive forces are computed which push the robot away from obstacles. Narrow passages are traversed with the support of a suction force which pulls the robot through them. Obstacles outside the field of view are kept in memory to prevent the wheelchair from rotating into them. The Obstacle Detection and Avoidance offers support for avoiding obstacles and traversing corridors. Further work is required to improve the required parameter set and the robustness of the algorithm so that it can be utilized in the daily life of a user. Different scenarios, like bypassing an obstacle, traversing a corridor and approaching a wall, are evaluated in simulation and on EDAN.