Águabots
Document register
- Poster1 pagePublished
- Presentation videoYouTubePublished
- Bill of materials39 KBPublished
- Team description paperNot shared
- Engineering journalNot shared
- Source code17.1 MB · 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
Our robot for RoboCupJunior Rescue Simulation 2026 is an autonomous rescue robot designed to explore a simulated maze, detect victims and wall tokens, avoid hazardous areas, build a semantic map, and return to the starting point within the competition time limit. The system combines a deterministic controller-side navigation strategy with ROS 2-based mapping and perception modules.
The movement system runs inside the Webots controller using the exploration_legacy library. It represents the arena as a discrete tile grid, reads LiDAR and ground-sensor information, selects safe neighboring tiles, executes calibrated wheel-position movements, and uses backtracking to return to the start. This approach was chosen because the Rescue Simulation maze is strongly tile-based and requires repeatable short movements rather than continuous generic path execution.
ROS 2 is used as a functional part of the final system for semantic mapping, victim and wall-token processing, visualization, debugging, and reporting support. The mapping module builds a tile-based semantic representation of the arena, while the vision pipeline uses ONNX neural networks and classical computer vision to detect victims and decode wall-token information.
The main strength of our system is the combination of competition-specific tile navigation, semantic mapping, and lightweight real-time perception in an integrated architecture.
Poster
Read the text of this document — 579 words
Wall and geometry mapping
TEAM
About us:
ÁGUABOTS
We are the Brazil team Águabots, representing the Farias Brito educational institution. The name INCHEON- ROBOCUP 2026 - RESCUE SIMULATION
Águabots was chosen to symbolize the fluidity, adaptability, and intelligence of our robots, just like the
importance of water for life and our mission to rescue victims in disaster situations, together with the BRAZIL
following members:
Isaac Newton S. Araújo Manuella A. N. Magalhães
Responsible for AI and Software Responsible for mapping
Archictecture
NAVIGATION AND EXPLORATION
Isabella F. G. P. Araujo Vinícius F.S. Oliveira
Custom tile-based exploration combining semantic mapping, LiDAR
Responsible for the design and Responsible for navigation and Exploration
path planning movement clearance analysis, backtracking and failed-goal memory.
Goal Selection Selects the next open unvisited neighbor tile or initiates backtracking
when no new tile is reachable.
Awards:
1st place in the Brazilian Robotics Competition (CBR) Planning DFS-style planning with weighted backtracking through visited non- Source: Authors, 2026
Rescue Simulation 2025. blocked tiles.
Motion Control Converts tile movements into calibrated Webots wheel-position
targets. VICTIM RECOGNITION
Safety & Cancels unsafe movements, reverses partial motion, marks blocked
HARDWARE Recovery tiles, and ensures return-to-start behavior. Image Processing
Frame filtering
Create a bitmask Duplicate detection removal
Use LIDAR to for the tile
detect walls
Object Detection
NO
Lightweight ONNX CNN
Depth-First Has the Are there
current tile any Breadth-First Is there an Detects: Target and victim
Search YES NO unvisited NO
Start already been unvisited Search (BFS) End
path
(DFS) visited? neighbors? available? Target Recognition
Source: Authors, 2026 Source: Authors, 2026
ROBOT COMPONENTS ROBOT IN SIMULATION CNN + Computer Vision
Move the Circle detection
Wheels: Two side-mounted wheels are responsible for the robot’s movement, allowing it to robot Polar unwrapping
move forward, make smooth turns, and rotate on its own axis. HSV color decoding
Update the tile YES
Inertial Unit: Used to ensure that the robot makes turns with exact 90º precision. Flowchart: Authors, 2026
bitmask Distance validation
Color Sensor: Used to identify the color of the simulated arena’s floor. It allows the detection
of holes that cannot be crossed. Victim Recognition
GPS: Used to record and transmit the precise location of victims detected by the robot. It is CNN Pipeline
also used to assist in mapping.
LiDAR: Used for accurate wall detection in the environment. The data provided by the LiDAR is MAPPING Bounding box detection
Image classification
essential to avoid collisions and to build representations of the traversed space. Fake victim rejection
The mapping system represents the Rescue Simulation maze as a semantic tile-based map, since the arena is
Cameras: Two cameras are positioned on the sides of the robot, assisting in precise victim
organized as a grid of square tiles Training & Deployment
detection through computer vision. Both cameras have a resolution of 64x64.
The mapper receives standardized ROS 2 data, mainly filtered odometry and LiDAR scans. Odometry provides the
Distance Sensor: Used to detect holes. It compensates for the color sensor’s limitation of robot’s position, while LiDAR detects nearby walls and obstacles. Ground sensors classify each tile as normal, colored,
analyzing only one point, making hole detection more reliable. swamp, hole.
The mapper converts sensor observations into a semantic tile representation
SYSTEM ARCHITECTURE of the maze.
WEBOTS CORE ROS 2 DECISION
MODULES
EKF TILE
LIDAR
LOCALIZATION EXPLORATION
TCP ROS
GPU + IMU BRIDGE SLAM PATH
TOOLBOX PLANNING
CÂMERA
CAMERA SEMANTIC TILE MOTION
MAPPER EXECUTOR
MOTOR
CÂMERA VICTIM MOTOR
COMMANDS
DETECTATION COMMANDS
Flowchart: Authors, 2026
Flowchart: Authors, 2026
Source: Authors, 2026
1 page, rendered as images so they load quickly. The text above is the document's own, extracted from the PDF.
Presentation video
Hosted on YouTube. The player loads only when you press play.
Bill of materials
Shown as the original PDF, because this one is smaller that way and its text stays selectable and searchable.
Source code
The team's own source code, 17.1 MB. It is a download rather than part of this page, because a zip is something you open on your computer. It comes from GitHub, which some school networks block.