Monarchy in The Swarm, Drone Swarm with Predetermined Leader

Published in

The Startup

7 min readJun 3, 2019

Drone swarms are groups of drones operating autonomously with their shared information. Autonomy is the key aspect here, they act as group, without remote control, relying on each other. They have their specific roles like surveillance, offensive actions and they also communicate with each other to complete their objective as swarm.

Their communication is over a wireless channel with limited range, they have low power and low general capability, they are prone to fail-more than traditional aircraft. All these properties of drones present an exciting problem in distributed systems domain.

Drone Types in the Swarm

For simplicity, we will assume there are two types of drones;

Leader Drone: A leader drone have extended range, high processing power and battery life. This drone will lead the actions of the swarm and may be used to connect swarm to other non-drone units in case of drone-assisted joint operation. A perfect example of this is, a military cargo plane dropping hundreds of drones over the contested area and only communicates with leader drone.
Follower Drone: Follower drones are less capable compared to the leader drone. These drones carry the burden of the mission, using their sensors to analyze battlefield. They have less communication range, battery, lifetime and processing power. They are expendable, unlike the leader drone.

Hierarchical Swarm Topology

Although there is the leader drone, it is not omnipotent. Operational area might be really large that drones would have to spread thin to cover as much as the operational area as possible. The leader drone might not have sufficient range to communicate with every drone in the swarm.

My proposal is multi-layer clustered drone swarm. Let’s see the roles of drones that can take while forming such swarm;

Leader Drone: This is the global leader drone. Its properties are given in the previous section.
Relaying Drone: Follower drones can be relaying drones. These are the helpers of the leader drone, effectively connecting the leader drone to the drones that are out of the range of the leader drone. When the leader issues a command, relaying drones carry the order to the whole swarm. The nodes who can not reach the leader directly sends their data to relaying drones instead for the same goal.
Edge Drones: Follower drones are the drones that are not relaying drones or the leader drone. They can be connected to the leader drone or to the relaying drone. Each edge drone can be connected to a single relaying drone at any given time.

Swarm Topology Discovery

Once the drones are over the contested area, they need to form a swarm quickly. This means, the leader drone should be connected to all other drones directly or indirectly. See the image below for the general problem.

How Can a Leader Drone Lead the Whole Swarm?

Let us first define three types of communication that is used in forming the swarm.

Lead Request: This will be used by the the leader drone. With this request, every drone within the range of the leader drone will recognize the leader’s existence.
Relay Request: This request will be used by the drones that are connected to the leader first. Drones that are not connected to leader but get this request will understand they are out of the range of the leader drone. They will connect to relaying drones instead. Depending on the size of the swarm, there might be multiple level of relaying. Every follower node broadcasts this request to cover as many of the free drones as possible to connect them to swarm.
Acknowledgement: These are messages that will be sent back after receiving lead request or relay request. If a drone sends acknowledgement request to leader request, it will connect to leader directly and will let leader know. Same things happen with relay request.

Example Topology Discovery

Let us assume physical placement of the drone is as given below;

L is, unsurprisingly our leader drone. We can see that L can not reach all of the drones in the future swarm. Let us see how our algorithm works;

L broadcasts Lead Request around it.
R and E1 receive Lead Request.
R and E1 accept L as swarm’s leader, send acknowledgement back.
L knows R and E1 is connected to it now.
Since E1 is part of the swarm now, it sends Relay Request to discover possible drones around it. Only L receives Relay Request but it ignores since it is the leader.
Since R is part of the swarm now, it broadcasts Relay Request to drones around it.
L hears R’s Relay Request but it ignores since it is the leader.
E2 receives R’s relay request. Since it has no leader at that point, it accepts the Relay Request of R. E2 will use R to connect to the leader L indirectly.
E3 receives R’s relay request. Since it has no leader at that point, it accepts the Relay Request of R. E3 will use R to connect to the leader L indirectly.
Since E2 is part of the swarm now, it sends Relay Request to discover possible drones around it. Only R and E3 receive Relay Request. R ignores it since it is the relayer of E2 already. E3 ignores it because it has R as relaying drone, providing shorter path to the leader drone.
Since E3 is part of the swarm now, it sends Relay Request to discover possible drones around it. Only R and E2 receives Relay Request. R ignores it since it is the relayer of E3 already. E2 ignores it because it has R as relaying drone, providing shorter path to the leader drone.

Connections in the Swarm at Steps 5 and 11

To summarize, leader starts topology discovery with Leader Request. Then, every follower of the leader broadcasts relay request to discover other drones who are not in range. Discovered nodes join to the swarm through relay nodes. Relay Requests will be repeated until whole swarm is formed. Therefore, some drones that are really far away from the leader may have to have multiple levels of relaying between them and the leader.

Advantages & Disadvantages of the Multi-Layer Clustered Architecture

At the end of our example topology, we can see L leads only to R and E1. We can also see R manages E1 and E2. We effectively created two cluster in a swarm. Instead of requiring every drone to be within the communication range of the leader drone, this architecture allows drone swarms to spread over large areas, as long as there is a drone which can relay its cluster’s data towards the leader.

Our example had one layer of relaying but there can be virtually infinitely more layers. This allows drones with low battery and less communication range to be a part of the swarm that can span multiples of a small drone’s communication range.

Clustering mechanism also eases management of the swarm. Imagine a leader drone that has to manage thousands of drones. Imagine a small drone can effectively relay the traffic of 10 other drones and leader drone that can manage 50 drones at any given time. With two layer of relaying in maximum capacity, we can virtually have a drone swarm that has 5550 drones(50 + 500 + 5000).

Breaking a swarm to clusters also allows cluster members to reduce the energy spent to transmit data over wireless channel. As we all know, wireless channel is shared and interference can happen. Less energy spent on a transmission results in less probability of interference with transmissions from other drones. Also, it results in less chance of detection of the drone swarms by anti-drone defenses. We can imagine this as a group of people whispering to their local leader and local leaders whispering to the global leader instead of whole population yelling at the global leader.

The Leader With The Help of Relay Drones Can Lead The Whole Swarm

The image above is a solution(with color mapping) of swarm control problem. Each colored area is under management of the drone with same color. Lines with same colors as drones are added to represent hierarchy in an easier to follow way. Although the leader originally has no communication range to cover the whole swarm, it can use relaying drones to reach whole swarm indirectly.

Drones are highly mobile. This means initial topology might quickly change. Developers must find the optimal point in recreating clusters(reclustering). Too many reclustering attempt can congest the data channel while too few reclustering can reduce effectiveness of the swarm.

Conclusion

Drone swarms are coming, they are already used in agricultural area. But those examples are swarm of 5 drones. Their military use is inevitable also. There are rumors of Chinese military using swarms of thousand autonomous drones as a swarm. And no, all those cool videos of drones forming beautiful pattern on a night sky is not example of autonomous swarms. Those were preprogrammed drones, they did not talk to each other at all.

Drone swarms create excellent problem in the domain of distributed systems. But their one significant advantage is that they are highly mobile entities, taking frequent topology change into account is a must.

The architecture I propose is no way unique or groundbreaking. Anyone with a knowledge of distributed systems and wireless computing would come up with the same or similar architecture. However, this architecture provides a good starting point for swarm applications.

Note: This architecture is based on my paper about Ad Hoc Networking in Sensor Networks. You can access it via the link below:

Recursive Clustering Time Division Multiple Access Scheduling