Jessica Vermeer
24 October 2019

Researchers from Delft University of Technology, the University of Liverpool and Radboud University Nijmegen have deployed a swarm of tiny drones to autonomously explore unknown environments. The work presented in “Science robotics” is a significant step in the field of swarm robotics. The drones weigh in at only 33 grams and need to navigate with limited sensing and computational capabilities. The scientists were inspired by the relative simplicity of insect navigation.

Credit: Guus Schoonewille, Delft University of Technology

The joint research was financed by the Dutch national science foundation NWO’s Natural Artificial Intelligence program. The goal was to take steps towards using swarm drones in search-and-rescue scenarios. After releasing a swarm for exploration, rescue workers can focus their efforts on the most relevant areas. Small drones can explore a disaster site much quicker than a single larger drone can.

“The biggest challenge in achieving swarm exploration lies at the level of individual drone intelligence,” says Kimberly McGuire, the PhD student who performed the project. “In the beginning, we focused on achieving basic flight capabilities such as controlling velocity and avoiding obstacles. After that, we designed a method for the drones to detect and avoid each other. We solved this by having each drone carry a wireless communication chip and then making use of the signal strength between these chips.”

Credit: Guus Schoonewille, Delft University of Technology

Limited sensing and computing of individual robots complicate navigation. Nature provided inspiration, as insects don’t make highly detailed maps. They just retain landmarks and behaviorally relevant places. “We reduced navigation expectations for the robots to just be able to get back to the base station,” says principal investigator Guido de Croon. McGuire adds: “The proposed navigation method is a novel type of bug algorithm. Bug algorithms do not make maps of the environment but deal with obstacles on the fly. This leads to less efficient paths than mapping but it can be implemented on tiny robots.”