UNSW VIP · AI4Everyone
We're building an AI-powered quadruped robot that navigates disaster zones, maps unknown environments, and identifies survivors — autonomously.
The Project
The RoboCup Rescue competition is an international robotics contest focused on developing robots capable of performing search and rescue operations in simulated disaster environments.
We equip a Unitree Go2 quadruped robot with advanced autonomous capabilities — environment exploration, victim identification, and real-time mapping — all running within the Robot Operating System (ROS 2).
Building on the efforts of previous teams, our goal is to create a highly competitive entry for the annual competition through sophisticated algorithms for SLAM, navigation, and computer vision.
What we build
Simultaneous localisation and mapping algorithms build a real-time 3D model of the environment as the robot moves through unknown terrain.
View detailsPath-planning and obstacle avoidance enable the robot to traverse complex, cluttered environments without any human input.
View detailsDeep learning models identify victims, hazards, and landmarks from camera and depth sensor streams, in real time.
View detailsAll modules — sensors, planners, controllers — communicate through ROS, enabling modular development and fast iteration.
View detailsLiDAR, IMU, and RGB-D cameras are fused to give the robot a robust, low-latency understanding of its surroundings.
View detailsBehaviour trees and reinforcement-learned policies coordinate exploration strategies to maximise victim-detection coverage.
View details
01 · Mapping
SLAM gives the robot a live understanding of where it is and what the environment looks like. The team works with LiDAR point clouds, map saving, coordinate transforms, and simulation pipelines so mapping can be tested before deployment on the Go2.
02 · Navigation
Navigation converts maps and mission goals into safe robot motion. This includes choosing reachable goals, planning paths around obstacles, recovering from blocked routes, and integrating movement commands with the quadruped control stack.
03 · Perception
Computer vision helps the robot recognise rescue-relevant objects from camera and depth data. The goal is to detect victims, hazards, and visual landmarks reliably enough to support autonomous inspection and reporting.
04 · System
ROS keeps the project modular. Sensors, mapping, navigation, control, web monitoring, and development tools communicate through topics, services, actions, and launch files so each subsystem can be developed and tested independently.
05 · Sensors
Sensor fusion combines LiDAR, IMU, camera, and depth information into a more stable estimate than any single sensor can provide. This improves mapping, localisation, and decision-making in cluttered rescue environments.
06 · Autonomy
Decision-making coordinates the robot's exploration strategy. It connects perception, mapping, and navigation so the robot can decide where to inspect next, when to recover, and how to prioritise rescue-relevant discoveries.
Project Progress
By May 2026 Achieved
Foundations & Simulation
By August 2026 In Progress
Hardware Integration
By End of 2026
Autonomy & Navigation
RoboCup 2027
Competition
External links
Official UNSW ChallENG project page with academic leads, application details, and team structure.
↗ CompetitionThe international RoboCup Rescue competition — rules, past results, and the global community we compete within.
↗ CommunityThe broader RoboCup initiative — robots competing in rescue, soccer, and home-assistance worldwide.
↗