AI-driven robot swarms evaluated to create larger and more efficient fleets
Researchers at the University of Southampton are investigating how to coordinate swarms of up to 100 autonomous vehicles that can work with a limited number of human operators.
Electronics and Computer Science’s Professor Sarvapali Ramchurn and Dr Danesh Tarapore are identifying the challenges posed by Artificial Intelligence (AI) algorithms automating robot swarms in a novel Pilot Project funded by The Alan Turing Institute.
The AI experts are working with industry partners at Thales and Dstl to shape the future of the technology’s design to enable swarms to have flexible autonomy in dynamic and uncertain environments.
Aerial, ground and underwater drones are being increasingly used in areas such as emergency response, ocean floor surveying, and product delivery, with operations currently relying on a human operator controlling one vehicle at a time.
In the coming years, it is anticipated that robots will need to be deployed in large numbers and with fewer operators to make the best use of their capabilities. Teams of operators might also collaborate to deploy their robots simultaneously for different objectives, such as fire and rescue services and non-governmental organisations responding to a natural disaster.
Professor Ramchurn, Director of the UKRI Trustworthy Autonomous Systems Hub, says: “AI algorithms have been developed to automate the actions of robot swarms in a cohesive and coordinated way. However, it has been shown that in some situations operators are overwhelmed or do not trust information coming from robots and therefore override them. By doing so, they may cause the system to fail. In other situations, completely relying on automation can mean obvious errors are not noticed in the system.
“To manage such large fleets in a safe manner, there need to be shifts in autonomy levels to allow humans to take corrective action. Understanding when such shifts should occur without losing out on the fault-tolerance benefits of a decentralised swarm, what levels of workload these shifts induce, and how the team of operators should enact such shifts are key questions that need to be addressed.”
The Turing-sponsored Pilot Project is developing the fundamental elements needed for research into the design of swarm coordination systems that can be flexibly controlled by human operators.
Previous work on supervisory control interfaces and multi-robot coordination has typically considered military applications with less than five drones and very simple scheduling models.
In contrast, the latest Southampton research is considering robots with more advanced coordination capabilities, such as modelling their environment, autonomously planning paths and allocating tasks to each other. These parameters should enable deployments of up to 100 robots with minimal human oversight.
One approach being evaluated for such swarms is for the robots to intelligently query the operator to sample and select suitable ‘plays’ designed to execute assigned sub-tasks. The swarm may also suggest characteristics of new plays depending on perceived changes in the environment, or seek to upgrade/downgrade the level of autonomy depending on the perceived uncertainty.