AURAT

Maze league · Italy · RoboCup 2026

Document register

Poster1 pagePublished
Presentation videoYouTubePublished
Bill of materials2 pagesPublished
Team description paper349 KBPublished
Engineering journalNot shared
Source code20 KB · GitHubPublished

Sharing each document is the team's decision. “Not shared” means this team chose not to publish it, or did not submit one — not that it is missing from the archive.

In their words

This paper presents the autonomous rescue robot developed by team AURAT for the RoboCupJunior Rescue Maze 2026 World Finals. Designed to achieve full exploration and precise victim identification within a strict 8-minute time limit, the robot features a robust, modular three-level physical architecture built with a custom vertical suspension system and hollow-rimmed TPU tires to safely navigate speedbumps and 25° ramps. The hardware ecosystem coordinates an Arduino Mega for low-level motion control, a custom 4-layer PCB for noise-free electronics routing, and a Raspberry Pi 5 acting as the high-level master node. To optimize computational efficiency, the software implements a concurrent, multi-threaded architecture utilizing Dijkstra’s algorithm for real-time dynamic pathfinding. Following technical challenges encountered during the national qualifiers in Catania, the vision pipeline was successfully upgraded from a blocking full-frame YOLOv8 model into an optimized, two-tiered framework. Spatial-chromatic blob localization is offloaded to an OpenMV H7 Plus edge camera, while character classification is executed on the Raspberry Pi 5 via a high-speed, rotation-invariant K-Nearest Neighbors (KNN) pipeline. System integration testing validates that these concurrent refinements guarantee deterministic obstacle navigation, robust cell alignment, and highly resilient target identification under variable arena conditions

Raspberry Pi 5 Arduino Mega OpenMV H7 Plus Dijkstra's Algorithm K-Nearest Neighbors (KNN)Hough Circle Transform Auto Centering PID EasyEDA PCB

Poster

Read the text of this document — 1094 words

CAP.MARCO MORETTI                               LORIS BERRA                                        SANDRA D’ARRIGO                      FEDERICO BARBAN
                                                      SOFTWARE ENGINEER                               ELETTRONIC                                        ARDUINO ENGINEER                        DESIGNER 3D

                                                                                                                            TEAM
                                          Our team brings together students from electronics and computer science, each contributing specialized skills that strengthen the project. Loris Berra, an
                                          electronics student with three years of RoboCup experience across multiple categories, has worked on industrial projects at STMicroelectronics and
                                          Pomini, and developed sensor‑based safety systems using ultrasound and YOLO‑based AI. He handled the electronic integration and PCB
                                          development. Marco Moretti, computer science student and Lead Software Engineer, specializes in Python, AI, mapping, pathfinding, and computer
                                          vision. He competed in the 2025 Italian Rescue Maze Championship, the 2025 OnStage Advanced World Cup - earning recognition for best technical
                                          documentation - and the 2026 Italian Rescue Maze Championship, finishing 5th. Sandra D’Arrigo, computer science student and Arduino specialist,
                                          competed in the 2025 and 2026 Italian Rescue Maze Championships and contributed to the visual design and presentation materials for the 2025
                                          World Cup. Federico Barban, computer science student and mechanical/3D modeling specialist, has been active in 3D design since age 12 and created
                                          all robot components from scratch. This is his first robotics competition. The strength of our team lies in the diversity of our skills - electronics, software,
                                          mechanics, and design. Each member brings a different perspective, and this variety allows us to solve problems more effectively, learn from one
                                          another, and continuously improve. Collaboration is essential: by combining our individual strengths, we transform complex challenges into achievable
                                          goals and build a more complete, innovative robot.

                     COMPONENTS                                                                       HARDWARE                                                                                                     INNOVATIONS
                                                                                                                        6

 1. 1× Raspberry Pi 5 (8GB): Handles high-level processing                                                     13                                   3     3
 2. 4× Motors: JGY‑370 DC 12V 100 RPM: Precise movement with
                                                                                                                                    7
    encoders
                                                                                   12
 3. 2× OpenMV Cam H7 Plus: For target & possible letter                                          7                                                                 7
    recognition                                                                                        1
                                                                                             1                                                                                       The robot uses 8 cm wheels equipped with a toothed TPU tread that ensures
 4. 1× Raspberry Pi Active Cooler: Cool down Raspy
                                                                                                                                                                                     excellent traction on ramps, obstacles, and uneven terrain. The tires are made
 5. 1× 9G Micro Servo Motor: Handles kits deployment                                                                                                                   2             entirely of TPU, a flexible and durable material chosen for its superior grip and
 6. 1× TCS34725FN RGB Color Sensor: Tile type recognition             2                                                         4
                                                                                                                                                                                     ability to maintain consistent contact with the ground. This combination improves
                                                                                                 14
 7. 6× VL6180X ToF Sensors: Mesure distance from walls                                                                                                                               stability during movement, enhances obstacle‑handling performance, and
 8. 1× BNO055 Gyroscope: For precise rotations & straight                                                                                                                            contributes to smoother, more reliable locomotion across irregular surfaces.
    movements                                                                                                               8
                                                                   18
 9. 1× Arduino Mega Dollatek PRO ATmega USB CH340G:                                                                                                        9
                                                                                                           7
    Manages low-level hardware operations
10. 1× TB6612FNG Motor Driver Board: Handle motors              15
                                                                                                                                        10                             2
                                                                                                 7                                                                 7
11. 1× PCB: Connect components while reducing cables                                    17

12. 1× XHC‑N 84 Power Bank: Charge the Raspy                              2
13. 1× MOD‑HW‑201 Infrared Sensor: Identify checkpoints
                                                                                                                                                              11                     Each motor is mounted on a dedicated suspension system that absorbs terrain
14. 1× HALJIA Red Push Button: Trigger a LoP
                                                                                                                                                                                     irregularities, reduces vibrations, and maintains consistent wheel‑to‑ground
15. 2× LED Strips: Illuminates the walls for better recognition                                                                     7
                                                                                                                                                                                     contact. The system helps the robot keep traction even when crossing uneven
16. 1x LED Strip: Report victims to the referee                               16
                                                                                                                                                                                     surfaces or small obstacles. The lubricated rods further minimize friction and
17. 1× Mini Toggle Switch: Cut the power
                                                                                                                    5                                                                mechanical oscillations, improving smoothness during acceleration and rotation.
18. 1× Sport Power 1600 mAh 11.1 V 120C Battery: Powers the
                                                                                                                                                                                     This setup increases overall stability, protects the mechanical components from
    motors & the LED strips
                                                                                                                                                                                     stress, and ensures more reliable movement in variable terrain conditions.

                                                                                                           SOFTWARE
                       HARDWARE MANAGING (ARDUINO)                                                                                                                     TARGET RECOGNITION (PYTHON)

                     COMMAND W

                                                                                                                                                   Leveraging a concurrent, multi-threaded software framework, Python and MicroPython form the core
     In our architecture, the Arduino Mega manages low-level hardware operations in real time, controlling the                                     cognitive backbone of our robot's high-level intelligence, seamlessly orchestrating real-time maze mapping,
     motors and polling sensors to guarantee immediate responsiveness. The only exceptions are the color and                                       dynamic pathfinding, and advanced computer vision.
     infrared reflection sensors, which are routed directly to the Raspberry Pi 5 to feed its high-level processing                                To achieve real-time execution, our vision pipeline was consist of a distributed, two-tiered framework.
     and strategic logic.                                                                                                                          Visual target localization is offloaded to an OpenMV H7 Plus edge camera running MicroPython. It isolates
     To ensure precise maneuverability, a full PID controller executes stable rotations by comparing target                                        regions of interest using wide-spectrum color blobbing followed by a localized Hough Circle Transform.
     angles with real-time feedback, while a wall-alignment routine dynamically calculates the robot's heading error                               The detected circle is split into five concentric rings analyzed via a LAB color space area-voting
     by applying the atan2 function to the differential readings of two parallel side-facing ToF sensors. finally a                                mechanism. By algebraically summing the numerical weights of the dominant colors, the system
     dedicated speed-ramping module smoothly governs acceleration and braking to prevent wheel slippage and                                        automatically identifies victim types and filters out fake targets. If a non-circular shape or character is found,
     minimize mechanical stress.                                                                                                                   the cropped frame is streamed to the Raspberry Pi 5 for letter classification.
     This integrated Master-Slave paradigm effectively balances deterministic hardware control with high
     computational power , ensuring highly reliable autonomous navigation even under complex arena conditions.                                                                             KNN (PYTHON)
                                       MAPPING (PYTHON)

                                                                                                                                                   Once a visual region of interest (ROI) is isolated by the OpenMV camera, the cropped frame is instantly
                                                                                                                                                   streamed to the Raspberry Pi 5 via a dedicated communication interface. A specialized Python-based K-
     The robot builds a map of the environment as it moves, updating it in real time using sensor and camera data.
                                                                                                                                                   Nearest Neighbors (KNN) pipeline then executes high-speed, rotation-invariant classification on the
     The area is divided into a grid of cells, where each cell stores information such as explored zones, tile types,
                                                                                                                                                   detected blob. By offloading the initial shape detection to the edge co-processor, the Raspberry Pi's CPU
     and walls. Python handles logging this data and updating the map with every forward step, marking features
                                                                                                                                                   overhead is minimized, dedicating its multi-threaded architecture strictly to classification and pathfinding. This
     encountered along the way. This enables the robot to avoid previously visited areas, choose more efficient
                                                                                                                                                   distributed approach ensures deterministic real-time processing, allowing the master system to instantly
     routes using Dijkstra algorithm, and plan the optimal path to the goal. Ultimately, this mapping system is an
                                                                                                                                                   update the maze navigation matrix or execute corresponding victim-handling routines without continuosly
     essential tool for autonomous navigation, allowing the robot to orient itself and make real-time decisions
                                                                                                                                                   halting the robot's exploration.
     based on exploration data.

        Modello 3D                                                                                                                           PCB                                                                                                           Github

Open as plain text

1 page, rendered as images so they load quickly. The text above is the document's own, extracted from the PDF.

Download the original PDF (7.1 MB) from GitHub