Intelligent Mobile Robot Controller Design for Hotel Room Service with Deep Learning Arm-Based Elevator Manipulator

Abstract

Although mobile robots have achieved great success in indoor navigation, they are still facing problems like operating through multi-floor hotel buildings. In this paper, a low-cost and lightweight mobile robot is proposed for hotel room service. For delivering any item that customers need, the robot should be able to manipulate elevators in an unmanned hotels. So, it is separated into three parts: elevator button detection, coordinate transform and fuzzy logical control to manipulate the robotic arm. To recognize and detect a variety of elevator panels and buttons, a large number of the images are collected and labeled manually by ourselves. With state-of-the-art deep learning framework, our model has achieved 95.172 mean average precision (mAP) even in elevators that are not included in training data. After properly detecting the elevator buttons, the 3D position corresponding to each elevator button is transformed by the fusion of two sensors which are a mono-camera and a Lidar. This paper presents a coordinates transform neural network to estimate real world position, and our method achieves in an average distance error of 1.356 mm. The fuzzy logical controllers are designed for manipulating the robotic arm fast and smoothly.