복합 재난 대응 시나리오를 위한 AI 기반 지능형 구조 로봇

목표 및 제작 동기

This project proposes a newly developed AI-based intelligent rescue robot service for complex disaster-response scenarios. The robot is designed to operate in simulated rescue environments where rescue targets must be detected, collected, transported, and released into designated safe zones under time pressure, limited visibility, physical obstacles, and possible external interference. The proposed system integrates K230 edge AI vision, STM32 embedded motion control, wireless remote communication, motor drive control, and a servo-based grasping mechanism. It can autonomously start, identify rescue targets and safe zones, approach targets using visual feedback, transfer the first rescue target to its own safe zone, and then continue operating with autonomous or remote-assisted control according to scenario requirements. The service goal is to provide a practical rescue-support platform for dangerous, narrow, or uncertain environments where direct human access is risky. [기대효과] The expected impact is to provide a low-cost and extensible robotic service model for early-stage disaster-response support. In real or simulated rescue scenarios, robots of this type can reduce direct human exposure to hazardous areas, improve the speed of locating and transferring small rescue targets, and provide a training platform for autonomous search, remote intervention, and multi-robot operation. The system also has educational and industrial value: it helps students and engineers practice integrated development across AI, embedded systems, mechanical design, and control algorithms. With further upgrades such as stronger obstacle avoidance, mapping, environmental sensors, and ruggedized mechanics, the system can become a practical platform for search-and-transfer missions in schools, factories, emergency drills, and public-safety demonstrations.

프로젝트 컨셉

The implementation uses a layered robot architecture. The perception layer runs an object-detection model on the K230 module to recognize rescue targets, color-coded safe zones, and priority target categories. The decision and communication layer exchanges compact serial commands between the K230 and STM32 controller: the STM32 sends task requests such as target search, safe-zone search, and priority search, while the K230 returns target ID and coordinates. The control layer uses PID-based visual alignment to convert target position errors into differential motor PWM values. The execution layer drives the chassis, controls the gripper servo, and performs pickup, transport, release, and retreat actions. A wireless communication module provides remote control when the scenario allows human assistance, while the autonomous start and first-target transport remain the core proof of unmanned operation. Safety mechanisms include stop states, manual override, bounded PWM outputs, collision-aware operation in a competitive field, and fail-safe behavior when no valid target is detected. [개발 역량] The team demonstrates end-to-end embedded AI robot development capability rather than simple kit assembly. The project includes non-standard mechanical design and manufacturing, K230 model deployment, STM32 firmware development, UART protocol design, motor PWM control, PID tuning, servo control, OLED debugging display, and wireless remote-control integration. The codebase contains separate scenario configurations, showing the ability to adapt the same robot platform to different safe-zone colors, target mappings, and rescue-field layouts. The vision program loads a KModel object-detection model, processes camera frames, filters target boxes according to safe-zone containment, selects targets by priority, and transmits structured results to the microcontroller. The STM32 program receives and parses the frame data, calculates control output, and drives the chassis and gripper through task-state conditions. This proves the team can connect AI perception, embedded control, electromechanical execution, and scenario-specific task logic into one functioning system. [지원 동기] The motivation comes from the practical difficulty of rescue work in constrained, chaotic, and potentially dangerous environments. Human rescuers often face risks from unstable structures, poor visibility, narrow spaces, and secondary accidents. The target scenario for this new project requires a robot to autonomously start, complete the first target transfer without remote intervention, protect itself from interference and collision, and move as many rescue targets as possible to the correct safe zone. These requirements closely resemble the core challenges of real rescue robotics: perception, decision-making, mobility, manipulation, safety, and robustness. The project goal is therefore to develop a clear service concept: an intelligent rescue robot that can identify targets, prioritize tasks, safely transfer objects, support remote assistance when needed, and provide a scalable foundation for future disaster-response applications.

팀원 기여도

• Junchi Xue4ds 회원
• Jun Hee4ds 회원
• Chenlin Yine4ds 회원
• Linzhen Daie4ds 회원