AUTO PATROL ROBOT

Challenges

        As the sole developer initiating the project, I navigated steep learning curves in ROS architecture and Autoware’s modular dependencies. Key tasks included physical sensor mounting and alignment, ROS node configuration for real-time data streaming, and validation of sensor fusion for localization accuracy. Initial attempts with Autoware Universe (ROS2) and Carla simulation were hindered by environment configuration issues and time constraints, forcing a pivot to Autoware.ai-1.14 in Docker for stability. Despite the lack of prior experience in autonomous systems, I led the transition from simulation to physical testing under tight deadlines, ensuring rapid progress in a resource-limited setting.

Challenges

        In the last days, my primary focus was on vehicle-level tasks, including map scanning and drawing, developing Autoware modules, implementing diverse docking strategies, and enhancing obstacle detection systems. Meanwhile, my colleague focused on testing new auto drive framework, optimizing system-level strategies, and setting up communication with the backend management system.

Auto Patrol

        A year ago, as a novice handling IMU and LiDAR for the first time, I could scarcely imagine achieving stable autonomous navigation in city parks within 12 months. Our small, inexperienced team tackled significant challenges in mastering Autoware's complex modules. From May 2025, we successfully get rid of Autoware and switched to self designed auto drive frameworks. While the system now operates reliably under standard conditions, edge cases still require refinement, and ongoing work - from sensor fusion to behavioral logic - remains under active development and optimization.

Application Service

        After achieving basic autonomous navigation, I led the development of application-layer algorithms, starting with a YOLO-based anomaly detection module for identifying behaviors like smoking and falling. Meanwhile, I integrated the alert systems, voice interaction, and vehicle tracking for targeted anomalies, alongside PTZ control for dynamic monitoring. Collaborating with frontend developer, we finalized a succinct voice interaction interface for enhancing user engagement.

User Experience

        The final user interaction and platform management experience were heavily influenced by various on-site constraints. Nearly half of the development time was spent in field testing, refining functionalities such as vehicle stability, obstacle avoidance, voice responsive sensitivity, and real-time video streaming. Early deployments always ended in failure, plagued by unexpected issues like loose cables, network latency, and minor hardware failures. Over months of iterative debugging and optimization, we finally achieved a relatively stable system.

Testbed Vehicle

        During a period of high chassis failure rates, our algorithm team urgently needed a stable test vehicle. Due to staffing changes in the mechanical and embedded departments, I independently designed a versatile test platform. This chassis featured four stepper motors for wheel direction control and two hub motor drivers for rotation. However, this project was temporarily paused due to the sudden running in robotic arm development. The stepper motors and their control algorithms were then repurposed for the 4-Axis Robotic Arm.

Future Plan

        This 4WD test vehicle, currently stationed at an unmanned seat right on my back, is scheduled for further development and testing in May, following the completion of the 4-Axis Robotic Arm project. This platform will serve as a critical tool for validating advanced navigation and perception algorithm in dynamic environments.