Flying robots to revolutionise disaster response by exploring buildings autonomously

An estimated 100 earthquakes worldwide cause damage each year. This damage includes collapsed buildings, downed electrical lines, and more. 

For first responders, assessing the scene and focusing rescue efforts can be critical and risky.

“A key idea of this research was avoiding redundancy in exploration,” says Seungchan Kim, a PhD student at Carnegie Mellon University’s Robotics Institute. 

“Since this is multi-robot exploration, coordination and communication among robots is vital. We designed this system so each robot explores different rooms, maximising the rooms a set number of drones could explore.”

The drones focus on quickly detecting doors because meaningful targets, like people, are more likely to be in rooms rather than corridors. 

To find these targeted entryways, the robots process the geometric properties of their surroundings using an onboard lidar sensor.

Gently hovering about 6ft from the floor, the aerial robots transform the 3D lidar point cloud data into a 2D transform map. This map provides the layout of the space as an image made up of cells or pixels, which the robots then analyse for structural cues that signify doors and rooms.

Walls appear as occupied pixels close to the drone, while an open door or passageway present as empty pixels.

The researchers modelled the doors as saddle points, which allowed the robot to identify passageways and pass through them quickly. When a robot enters a room, it appears as a circle.

Kim explains that the researchers opted for a lidar sensor over a camera for two main reasons. First, the sensor uses less computing power than a camera. 

Second, conditions inside a collapsed building or at the site of a natural disaster might be dusty or smoky, which would impair the vision on a traditional camera.

No centralised base controls the robots. Rather, each robot makes decisions and determines optimal trajectories based on its understanding of the environment and communication with the other robots. 

The aerial robots share the list of doors and rooms they’ve explored with one another and use this information to avoid areas that have already been visited.

The team presented their research last month at the 2024 IEEE International Conference on Robotics and Automation.

Kim plans to continue researching multi-robot exploration and semantic scene understanding for multi-robot exploration and task coordination.