Building smarts into drones
Defence researcher Robert Hunjet runs the Self-organising Communications and Autonomous Delivery Service (SCADS) project.
Through SCADS he is proposing an unmanned autonomous system that will allow critical information exchange to and from Australian warfighters in times of communication stress.
To prove the SCADS concept, Hunjet and his team at DST are using octocopters that fly themselves.
One criticism of current military drone usage is the existence of a control channel which can be jammed, causing the drone to return to base.
“If the decision-making and autonomous flight is embedded in the octocopter, then it can manoeuvre itself in a way that best achieves the mission goal,” says Hunjet.
“We are looking at having an onboard decision-making process that determines how it should move, given it has limited knowledge of surrounding components, so that the behaviour of the swarm is intelligent.”
Hunjet is inspired by the flocks of hundreds of starlings that move around the sky without crashing into each other.
“If we were to send information to each of a 500 piece octocopter swarm, telling them where they should be, we would flood the network,” he says.
“We don’t want to use the network to coordinate the swarm of Unmanned Aerial Vehicles.”
From a SCADS perspective, local rules are needed to be specified on each octocopter to give the swarm the ability to pass critical information between warfighters in a highly contested environment.
“The power of swarm robotics and emergent behaviour is that we are talking simple implementation and low power on cheap platforms to give intelligent results,” explains Hunjet.
“Think high-tech courier pigeons.”
Hunjet says a SCADS swarm needs to learn what should be done in different contexts. In order to be able to gather information from the environment and control the unmanned systems in a platform-agnostic manner, DST has created software called Hardware Abstraction and Integration Layer (HAIL).
This software allows the system to perceive the environment, and modify the behavior of the Unmanned Aerial Vehicle.
Machine learning algorithms will be run with simulations offline so the units understand what they should do in certain situations, much like the pre-mission training that our troops undertake before deployment. That is combined with a real-time feedback loop that allows SCADS elements to tune responses on the fly.
As an initial, readily implementable proof of concept, SCADS is exploring a stigmergy-based data ferrying approach. Stigmergy is the process of communicating through the marking of the environment.
The DST developed algorithm enables autonomous control of drone speeds to facilitate information exchange between disconnected network nodes.
“We are now planning a three year program with the US Navy Postgraduate School that will include progressively more complicated swarming trials in California and Woomera,” says Hunjet.
SCADS also has research agreements with several Australian universities.
“The support we’ve received from the ADF and academia indicates that we are on the right track.”