Robot Following System Based on Convolutional Fast Sliding Window Approach for Motorized Wheelchair

Commercially available motorized wheelchairs do not have the following function, which causes great inconvenience to patients wearing exoskeleton robots for rehabilitation training. We propose an exoskeleton robot following system for motorized wheelchairs based on a convolutional fast sliding window approach. The proposed system tracks the robot’s motion in real-time and makes predictions about its movement path. More specifically, a pre-trained neural network is responsible for localizing the robot’s location in the camera’s coming frames. The position information is post-processed and converted into hardware commands that control the operation of the wheelchair motors. The motorized wheelchair then moves according to the received commands. We implemented our system on the hardware of an embedded system and evaluated our approach in the natural environment. Experimental results in both indoor and outdoor environments show that the proposed method can accurately localize and follow a moving robot in real-time.