Who are We? wall token detection We are from Argentina, participating for the second time in the Uses a custom image processing method to detect victims and international RoboCup competition, in the Rescue Simulation category. cognitive targets on walls, using the front and left cameras. A color Previously, we took part in RoboCup Americas 2025 In which we got the threshold highlights key colors and removes background noise, then first place and then in Robocup Salvador 2025 in wich we got third place. contours are extracted to identify potential signs. Victims are recognized This poster presents the most important aspects of our robot's software, by analyzing black and white pixel distribution across five regions in a 3×3 highlighting the strategies and improvements developed for RoboCup 2026. grid. Cognitive targets are confirmed through shape and color validation — near-square contour, minimum area, valid color present — with no letter team: analysis needed. Martina Talamona Ramiro Francavilla Emanuel Hamui Programming & testing. Programming & testing. Mentor. Original Proper Concentric Original Proper Cleaned image threshold coloured rings image threshold image robot’s hardware components Three 40x40 cameras (left, front and right) for robust visual analysis, enabling the detection of signs, fake tokens and Cognitives targets. A differential drive system with two independently controlled wheels navigation system for precise movement and smooth turns. We use a hybrid pathfinding system combining Dijkstra’s algorithm A distance sensor for obstacles detection aided by Lidar. for global planning and Greedy Best First Search for precise, pixel-level A LIDAR sensor provides high-resolution 2D mapping and accurate navigation in tight spaces. The map grows dynamically with LiDAR data, obstacle detection, improving path planning when combined with adapting to any environment size. Obstacles like walls and black holes camera system. are marked as non-traversable. Swamp tiles are assigned an additional GPS: due its noise the Gps is still being fundamental, the robot traversal cost, discouraging re-entry while still allowing passage if no averages multiple readings while stationary and applies stronger other route exists. Straight paths are prioritized to optimize movement, corrections after a LOP event to maintain a stable position estimate. straight paths are prioritized, and the final route is smoothed to Inertial unit: For directional alignment. eliminate unnecessary steps. We use Godot Engine to create a real-time visualization. This allowed us to improve our navigation algorithm. robot’s software visualizer (godot engine) We use Python as our programming language. Our code follows a modular structure with over 10 specialized modules for tasks In the visualizer we can see such as navigation, mapping, image processing,Vector for the cost of each minitile(g) mathematical functions, tile classification, and robot control. and the number of walls(w) This design simplifies maintenance, testing, and expansion by it has around it. assigning clear responsibilities to each subsystem and supporting Points painted as violet are efficient team collaboration. the lowest cost path that the robot has planned. Black color represent walls innovations and obstacles. We split obstacle detection using Lidar and the front camera; Lidar to see if there is something close, and as a helper is the front camera, we take the left most, the right most and the center pixel and we compare those values exlcuding walls and token colours. Once the mapping obtacle is confirmed, they are marked as "X" in the matrix Our mapping system constructs a dynamic, grid-based map as the robot Robocup Salvador 2025 third place explores. Each tile is subdivided into four sections to allow precise classification representation. Robocup Americas 2025 first place We also implemented a swamps avoidance navigation, where each of terrain types and obstacles. LIDAR sensors detect walls and continuously detected swamps pixel receives an additional cost, and it’s minitile update the map in real time. As the robot moves, the map expands automatically visit count is increased proportionally to further discourage re-entry. to include newly discovered areas. Tiles are classified using sensor input, We implemented a positioning based in the LiDAR’s front ray (ray distinguishing features such as swamp, checkpoint, or black hole. Obstacles are 256), each time the robot moves, the difference between the previous marked as non-traversable and recorded as "X" in the matrix representation, while visited tiles are displayed in green. For efficient pathfinding and local analysis, the CAETI-UAI and current front-ray distance is used to calculate how far it traveled in X and Y based on its heading. system maintains a cropped matrix centered around the robot’s current position. example of Map We fail. We learn. We improve. We try again. matrix how we That is our principle. with represent technologies the obstacles: Map: