In a case of science mimicking nature, engineers since the late 1980s have applied artificial intelligence to create systems of robots and drones that form swarms, or groups that communicate with each other to complete a shared task, similar to those found in certain insect groups, birds or even bacteria. While engineered swarms have been of interest to engineers for many years for tasks such as search and rescue and military operations, they often require full information of their target’s position and trajectory, which is not always available in real-world situations. They also can only be completed while the swarm remains in certain formations such as a circle, another condition that hinders their effectiveness.
Now, a team of researchers led by Antonio Bono, doctoral student at the ICT School of the Department of Informatics, Modeling, Electronics and Systems Engineering at the University of Calabria in Italy, has developed a method for allowing swarms to capture a target in an ellipsoidal region even when previously required information, such as target’s velocity and acceleration, is absent.
Researchers proposed a time-dependent network control model based on the characteristics and behaviors of temporal networks. Results indicate a promising strategy to tackle large, complex networks.
A. Bono, L. D’Alfonso, G. Fedele, and V. Gazi, “Target capturing in an ellipsoidal region for a swarm of double integrator agents,” IEEE/CAA J. Autom. Sinica, vol. 9, no. 5, pp. 801–811, May 2022.
In many swarm scenarios, the leader of the swarm may use computer vision, which is computationally expensive and therefore not applied to all agents, in order to track the target, while the followers use proximity sensors such as ultrasonic or infrared sensors, to track the other swarm agents near them. Because of this, it is common for agents to have an awareness of each other, which is critical for a successful mission, while only having uncertain knowledge about the target.
For the study, the agents in the swarm were modeled as double integrators, which is another name for point mass dynamics describing motion based on Newton’s second law of force equals mass times acceleration, according to Veysel Gazi, professor of electrical and electronics engineering at Yildiz Technical University in Turkey and a senior member of the research team.
“It is a generic model which can be used to describe the dynamics of many different systems such as robots, drones and so on,” he said.
In the scenarios modeled by the researchers, each agent knew the relative position of the target and had an estimation of its velocity and acceleration. As such, the swarms are able to use an estimate of the target’s position until they obtain the exact information.
With estimation errors being bound by known values — that is, with knowledge of the range within which the target must be—the researchers were able to design a control law that ensured that the agents enclosed on the moving uncertain target.
A second significant choice in the work is the shape of the area around the target occupied by the agents.
“In our work, capturing/enclosing [a moving target, such as an adversarial intruder,] over an ellipsoidal region — in contrast to circular regions — is considered for the first time,” said Gazi. “Moreover, uncertainties about the target motion in terms of knowledge of its velocity and acceleration are considered in the paper and the mission is achieved under these uncertainties.”
“Ellipsoid is a shape which requires different level attractions/repulsions on different semi-axes,” Bono said. “Differently from many other approaches, the ellipsoidal ring constitutes a non-trivial geometric constraint to address because of its non-convexity.”
The researchers successfully demonstrated this work with two numerical simulations and a software in the loop simulation.
“Because the work utilizes a generic agent model and realistic assumptions, the results can be implemented on many engineering swarms composed of different type of agents,” Gazi said. “[Potential applications include] target/intruder tracking/capturing, patrolling areas or escorting missions by autonomous agents such as autonomous ground, sea, or air vehicles.”