02 June 2023
Swarms robots have potential in the real-world applications in logistics, search and rescue, construction, space exploration, and many more. However, although robust through redundancy of a large number of individuals, each robot is susceptible to failure at any given time, which may decrease the performance of the swarm as a whole. Hence, to take these applications from a research environment to real-world implementation, it is necessary to mitigate such faults. The difficulty with designing a good mitigation strategy lies in the complexity of the swarm as a system, where individual interactions give rise to emergent behaviour.
In this work, EMERGE partners from the University of Bristol aim to learn a mitigation strategy using neuroevolution in a fault-discriminatory metric space and demonstrate the strategy in a realistic intralogistics use-case.
The authors focus on two variations of an intralogistics task, selected to test how mitigation strategies might vary with the interdependence of members of the swarm: random walk behaviour for box retrieval and delivery, and collective transport where multiple robots must coordinate to do the same. The group was able to extract metrics based on local-sensing which were highly indicative of a fault. By learning strategies on this metric space, which has high discriminatory power between faulty and normal states of a robot, we can reduce the search space of strategies.
Read the paper in the link below.
More Information
