An Autonomous Service Mobile Robot for Indoor Environments

Abstract

This paper proposes an autonomous indoor service robot system framework integrated with multi-sensors to implement the service robotic functions. Both the voice recognition and real-time object detection capabilities are incorporated in this system. Firstly, a novel mapping strategy is proposed to build a hybrid grid-topological-semantic map which is used for voice-control-based navigation for robots. Secondly, in order to reduce the power consumption of mobile robot object detection, a low-power parallel acceleration method using neural network accelerator sticks is proposed. And the power consumption is reduced by 15 times with approximately the same frame rate compared with laptop GPU GTX 960M. Moreover, the YOLO model used in this system is about 10 times faster than two-stage methods. In addition, the experiment result demonstrates that the proposed mapping strategy can be well combined with the voice recognition system to achieve voice-controlled navigation. The structure of the robot system we designed in this paper can be widely applied to indoor service robots.