Distilling Tiny and Ultra-fast Deep Neural Networks for Autonomous Navigation on Nano-UAVs

TL;DR

This paper introduces a highly compact and ultra-fast neural network for nano-UAV navigation, significantly reducing memory use while maintaining high accuracy and success in complex obstacle scenarios.

Contribution

It presents a novel CNN distillation method that reduces memory footprint by up to 168x and improves inference speed, enabling effective autonomous navigation on nano-UAVs.

Findings

Tiny-PULP-Dronet v3 achieves 100% success in complex navigation tasks.

Memory footprint reduced by up to 168x compared to original PULP-Dronet.

The new CNN outperforms state-of-the-art in real-world nano-UAV navigation scenarios.

Abstract

Nano-sized unmanned aerial vehicles (UAVs) are ideal candidates for flying Internet-of-Things smart sensors to collect information in narrow spaces. This requires ultra-fast navigation under very tight memory/computation constraints. The PULP-Dronet convolutional neural network (CNN) enables autonomous navigation running aboard a nano-UAV at 19 frame/s, at the cost of a large memory footprint of 320 kB -- and with drone control in complex scenarios hindered by the disjoint training of collision avoidance and steering capabilities. In this work, we distill a novel family of CNNs with better capabilities than PULP-Dronet, but memory footprint reduced by up to 168x (down to 2.9 kB), achieving an inference rate of up to 139 frame/s; we collect a new open-source unified collision/steering 66 k images dataset for more robust navigation; and we perform a thorough in-field analysis of both PULP-Dronet and our tiny CNNs running on a commercially available nano-UAV. Our tiniest CNN, called Tiny-PULP-Dronet v3, navigates with a 100% success rate a challenging and never-seen-before path, composed of a narrow obstacle-populated corridor and a 180{\deg} turn, at a maximum target speed of 0.5 m/s. In the same scenario, the SoA PULP-Dronet consistently fails despite having 168x more parameters.