Tiny-PULP-Dronets: Squeezing Neural Networks for Faster and Lighter Inference on Multi-Tasking Autonomous Nano-Drones

TL;DR

This paper presents Tiny-PULP-Dronet, a highly compressed neural network that enables faster, lighter, and multi-tasking capable autonomous nano-drones with minimal performance loss.

Contribution

It introduces a novel model compression methodology that reduces model size by over 50 times and computational complexity by 27 times, enabling advanced onboard autonomous navigation.

Findings

50x fewer parameters than previous models

27x reduction in multiply-and-accumulate operations

Maintains similar flight performance

Abstract

Pocket-sized autonomous nano-drones can revolutionize many robotic use cases, such as visual inspection in narrow, constrained spaces, and ensure safer human-robot interaction due to their tiny form factor and weight -- i.e., tens of grams. This compelling vision is challenged by the high level of intelligence needed aboard, which clashes against the limited computational and storage resources available on PULP (parallel-ultra-low-power) MCU class navigation and mission controllers that can be hosted aboard. This work moves from PULP-Dronet, a State-of-the-Art convolutional neural network for autonomous navigation on nano-drones. We introduce Tiny-PULP-Dronet: a novel methodology to squeeze by more than one order of magnitude model size (50x fewer parameters), and number of operations (27x less multiply-and-accumulate) required to run inference with similar flight performance as PULP-Dronet. This massive reduction paves the way towards affordable multi-tasking on nano-drones, a fundamental requirement for achieving high-level intelligence.