Pavel Petráček presents Bio-Inspired Compact Swarms of UAVs without Communication and External Localization

On 2020-12-17 11:00:00 at The seminar will be held online
In the talk, we will present and discuss our preliminary results in deploying a
fully decentralized swarms of aerial robots in the real world using onboard
visual relative localization in the position control feedback loop of each

The seminar will be via Zoom:
Meeting ID: 990 4871 6592
Passcode: 696653

This article presents a unique framework for deploying decentralized and
infrastructure-independent swarms of homogeneous aerial vehicles in the real
world without explicit communication. This is a requirement in swarm research,
which anticipates that global knowledge and communication will not scale well
with the number of robots. The system architecture proposed in this article
employs the UltraViolet Direction And Ranging (UVDAR) technique to directly
perceive the local neighborhood for direct mutual localization of swarm members.
The technique allows for decentralization and high scalability of swarm systems,
such as can be observed in fish schools, bird flocks, or cattle herds. The
bio-inspired swarming model that has been developed is suited for real-world
deployment of large particle groups in outdoor and indoor environments with
obstacles. The collective behavior of the model emerges from a set of local
rules based on direct observation of the neighborhood using onboard sensors
only. The model is scalable, requires only local perception of agents and the
environment, and requires no communication among the agents. Apart from
simulated scenarios, the performance and usability of the entire framework are
analyzed in several real-world experiments with a fully-decentralized swarm of
unmanned aerial vehicles (UAVs) deployed in outdoor conditions. To the best of
our knowledge, these experiments are the first deployment of
decentralized-inspired compact swarms of UAVs without the use of a communication
network or shared absolute localization. The entire system is available as
open-source at

Standard seminar length ~ 40 min talk, 20 min discussion

Link to paper:

Responsible person: Petr Pošík