Evolving hardware instinctive behaviors in resource-scarce agent swarms exploring hard-to-reach environments

This work introduces a novel adaptation framework to energy-eciently adapt small-sized circuits operating under scarce resources in dynamic environments, as autonomous swarm of sensory agents. This framework makes it possible to optimally congure the circuit based on three key mechanisms: (a) an o-line optimization phase relying on R2 indicator based Evolutionary Multi-objective Optimization Algorithm (EMOA), (b) an on-line phase based on hardware instincts and (c) the possibility to include the environment in the optimization loop. Specically, the evolutionary algorithm is able to simultaneously determine an optimal combination of static settings and dynamic instinct for the hardware, considering highly dynamic environments. The instinct is then run on-line with minimal on-chip resources so that the circuit eciently react to environmental changes. This framework is demonstrated on an ultrasonic communication system between energy-scarce wireless nodes. The proposed approach is environment-adaptive and enables power savings up to 45% for the same performance on the considered case studies.