In this paper, we propose an autonomous stair-driving system for the stable traversal of stairs by a tracked mobile robot operating in indoor disaster environments. Before developing the system, we conduct dynamic simulations to analyze the requirements for the robot to climb stairs. Simulations are performed under various initial conditions, and based on a detailed analysis of the results, we derive the necessary conditions for the robot's ascent. Using these requirements, we design the autonomous stair-driving system, which includes three main components: stair approach, stair alignment, and stair traversal. First, during the approach stage, we present a strategy for recognizing stairs using an object detection algorithm and generating control inputs for the stair approach motion. Next, in the alignment process, we outline an image processing sequence that extracts the edge contour of the stairs and a method for generating control inputs from the combined contour. Finally, in the traversal sequence, we describe the strategy for driving up the stairs. Additionally, we introduce an integrated ROS system to ensure the sequential execution of each strategy. We also verify the effectiveness of the individual strategies and demonstrate the capability of the proposed system through experiments using mock-up stairs and tracked robots.

In this study, a module combining various types of sensors was developed to increase search efficiency inside collapsed buildings. It was designed to be less than 70 mm in diameter so that it can be put into narrow spaces, and is equipped with a small & high-performance processor to process multiple sensor data. To increase sensor data processing efficiency, multi thread based software was configured, and the images were combined and transmitted to ensure time synchronization of multi-channel video data. A human detection function based on sound source detection using two microphones was implemented. The developed multi-sensor module was tested for operation by mounting it on a snake-type robot in a test bed simulating a disaster site. It was confirmed that the visible range of the robot to which the multi-sensor module was applied was expanded, and the ability to detect human and low-light human detect was secured.

In this study, we proposed microphone array and algorithm for sound source localization based on GCC-PHAT for the robot searching victims in a narrow space. Through frequency domain analysis, we designed filter to make algorithm react only to the sound with a human voice frequency. Additionally, calibration algorithm was integrated to solve the problem of the update cycle of result value becoming very short when passing through the filter, presenting difficulty in checking the value. Results obtained through experiments verified the performance of the proposed microphone array and sound source localization algorithm.