SOFTWARE TEAM Line-following Evacuation Zone The team was founded in 2023 under the original name 'Unicorno colorato' (which translates to 'Colored Unicorn'). The initial team members were: Simone Bettega, Matteo Defant, Samuele Contessotto, and Luca Facchinelli. Over the To follow the line, we implement a PID For object recognition inside the evacuation years, the team underwent several lineup changes (some team members and algorithm, a negative feedback system room, we implemented an integrated friends can be seen in the 'nazionali 2025' image). In 2024, QuantumSpeed was widely used in automatic control systems. computer vision pipeline. Sphere detection established with the goal of breaking into the national top 10. That year, we We detect the black line using an HSV is performed using a TensorFlow Lite achieved a second-place finish, qualifying for the European Championship in threshold and calculate its center of mass. machine learning model, trained in Google Bari (IT), where we reached the podium again in second place. This year, we Next, we determine its distance from the Colab on a dataset of approximately 3,000 decided to completely overhaul the robot's underlying technology, transitioning camera's center. Based on the difference images. Triangle recognition, on the other from traditional sensors to a camera-based system. This choice led us to win the along the x-axis, the PID function calculates hand, relies on an HSV mask–based regional competition, secure second place at the national level, and qualify for the required velocity variation for the wheels. algorithm developed with the OpenCV the World Championship in Corea, which will mark our final chapter in the RCJ. library. Image acquisition is handled by the This variation is then multiplied using the Simone Bettega Pi Camera Module 3 Wide, which provides a difference along the y-axis (see image); this AI model trainer and sistemistic expert 120° field of view and real-time autofocus approach allows us to follow the line He developed and implemented a machine learning At the beginning of this year, we decided to switch from sensors to capabilities. The entire system is smoothly and reliably. system for silver line detection. He also implemented the a camera setup. To do this, we introduced a Raspberry Pi 5 8GB programmed in Python, and its We also detect other colors using HSV performance is improved by the AI HAT+ lack-of-progress recovery procedure using file service. In equipped with two Pi Camera 3 Wide lenses. We chose the Raspberry Pi because it is highly versatile, user-friendly, and thresholds. When red is detected, the robot accelerator delivering up to 13 TOPS of 1st Place addition, he collaborated on improving the PID control Qualifying Tournament - Trento 2026 well-suited for a wide range of applications. Furthermore, we stops instantly, while green tells it which computing power, enabling inference system, contributing to more accurate and stable robot integrated a hardware accelerator to achieve even higher direction to turn. To recognize intersections, speeds of around 30 FPS and ensuring navigation. He also took care of creating the venv. performance.The Raspberry Pi handles both line-following and we also utilize concentric circles: if they highly reliable and efficient object victim recognition. To run the entire system seamlessly, we set up intersect the line more than twice, an recognition. Finally, a specially designed Luca Facchinelli a dedicated environment and configured service files, allowing the intersection is detected. Finally, to recognize collection chest allows the robot to gather all Co - Captain, Line following programmer code to execute autonomously. the silver line, we employ machine learning three spheres simultaneously, significantly 2nd Place Luca has been part of the team since the beginning. and calculate the difference in reflected light reducing the overall evacuation time and nd 2 Place European Robocup - Bari 2025 Initially, his role was to coordinate the team, leveraging his with the LEDs turned on and off. improving task efficiency. Italian Open - Catania 2026 prior experience in RoboCup Rescue Line. Over time, he Q UANTUM S PEED specialized in C++ programming for Arduino, and over the past year, in Python for the Raspberry Pi. Currently, his primary responsibility is programming the line-following subsystem. The entire program is built on Python, a versatile and flexible programming language known for its ease of use. Python was Matteo Defant selected due to its widespread adoption, particularly in image Captain, evacuation zone programmer and 3d modelling processing applications. The rich ecosystem of libraries that Defant, the team captain, was primarily responsible for Python offers was a decisive factor in our choice to use it for programming the evacuation zone. Since 2025, he has our robot. Libraries such as OpenCV and NumPy contribute significantly to our code's functionality, efficiently handling “DA VINCI” HIGH SCHOOL (TN) - ITALY RESCUE LINE developed and implemented a machine learning system to recognize and classify the victims. In addition, he heavy computational workloads. designs the robot in 3D using Fusion 360 and has taken care of its assembly. HARDWARE Chassis and wheels Rescue mechanism 4x SERVOMOTORS 6x MICROSWITCHES Two are used to detect 2x DC-DC CONVERTERS Step Down ADAFRUIT LSM6DSOX+LIS3MDL 9g whether the robot has One step-down converter is A sensor used to measure MULTIPLEXER The victims collection They close the gripper collected the victim, while used to step down the battery 3-axis inclination. It is utilized An I2C multiplexer, required arms of the collection the others are used to voltage from 7.4V to 5V for the for detecting slopes and to expand I2C addresses and system was entirely mechanism and open prevent the robot from Raspberry Pi, while the other ramps, as well as for executing communicate with multiple designed and 3D-printed the exit channels. getting stuck. powers the servomotors. more accurate turns. sensors simultaneously. They offer an by us. The mechanism excellent power to consists of a ripper with weight ratio. two independent arms, so RASPBERRY PI 5 2x OVONIC LiPo BATTERIES that when a victim is 8GB We use two batteries: the collected, it is then guided Single-board first, rated at 5200mAh and into the correct channel. computer; after several 7.4V, powers the Raspberry tests, we opted for this Pi, while the second, rated at Since our first partecipation, we decided to Inside each arm, there is solution due to its 2200mAh and 11.1V, powers a microswitch to detect whether the victim has versatility and the motors and sensors. custom-build the entire robot ourselves, with the computing power. exception of the sensors. The chassis was 3D been caught. Additionally, there is a circuit that, 2x PICAMERA printed in PLA using a Bambu Lab P1P, while The when closed, identifies whether it is a silver, MODULE 3 WIDE DC-DC CONVERTER wheels have a TPU inner skeleton with a silicone black, or fake (gray) victim. Step Up Two wide-angle coating . Additionally, all wiring and circuit boards The bottom of both cameras to achieve the One step-up converter is maximum field of view used to step up the battery were custom-made. This approach allows us to channels is closed by a of both the line and voltage from 11.1V to 12V for position all components exactly where we prefer. rod connected to two the evacuation zone. the motors. We placed two cameras at the front of the robot, other servomotors, one dedicated to line following and the other for which open only when LOGIC LEVEL victims recognition, ensuring that everything fits the robot reaches the SHIFTER RASPBERRY PI AI HAT+ together perfectly. evacuation zone. In total, Used to enable A graphics accelerator used the mechanism consists communication in the victims recognition between the Raspberry phase to increase speed and of five servomotors and Pi (3.3V) and the LEDs reliability. is printed using Bambu (5V). Lab's PLA Basic. The two arms are also highly 2x LED STRIP practical for separating victims if they happen to We use two RGBW 4x GEARMOTOR POLOLU be close to each other, preventing the risk of LED strips to ensure SERVOMOTOR 25kG 5x TOF SENSORS We use four Pololu motors THANKS TO: consistent lighting, It raises and lowers the SERVO DRIVER They are used to with a 156:1 gear ratio, collecting two at once instead of one. making the system collection mechanism, Required to control the bypass obstacles powered at 12V to ensure independent of providing good accuracy servomotors directly along the path and to both speed and power. external illumination. and reliability. from the Raspberry Pi. navigate within the evacuation zone.