This website requires JavaScript.

Mantis: Enabling Energy-Efficient Autonomous Mobile Agents with Spiking Neural Networks

Rachmad Vidya Wicaksana PutraMuhammad Shafique
Dec 2022
Autonomous mobile agents such as unmanned aerial vehicles (UAVs) and mobilerobots have shown huge potential for improving human productivity. These mobileagents require low power/energy consumption to have a long lifespan since theyare usually powered by batteries. These agents also need to adapt tochanging/dynamic environments, especially when deployed in far or dangerouslocations, thus requiring efficient online learning capabilities. Theserequirements can be fulfilled by employing Spiking Neural Networks (SNNs) sinceSNNs offer low power/energy consumption due to sparse computations andefficient online learning due to bio-inspired learning mechanisms. However, amethodology is still required to employ appropriate SNN models on autonomousmobile agents. Towards this, we propose a Mantis methodology to systematicallyemploy SNNs on autonomous mobile agents to enable energy-efficient processingand adaptive capabilities in dynamic environments. The key ideas of our Mantisinclude the optimization of SNN operations, the employment of a bio-plausibleonline learning mechanism, and the SNN model selection. The experimentalresults demonstrate that our methodology maintains high accuracy with asignificantly smaller memory footprint and energy consumption (i.e., 3.32xmemory reduction and 2.9x energy saving for an SNN model with 8-bit weights)compared to the baseline network with 32-bit weights. In this manner, ourMantis enables the employment of SNNs for resource- and energy-constrainedmobile agents.