Smart Drones: SWaP-C Analysis for Agentic Drone Operations
Name
Russell_Final.pdf
Size
3.44 MB
Format
Adobe PDF
Checksum (MD5)
523742cc011be61921937b660d38ed1a
Author(s)
Russell, Holt
Wheelhouse, Brian
Russell, Thomas
Kokotajlo, Timothy
Carpenter, Dylan
Che, Dylan
Russell, Joanna
Spellmann, Benjamin
Hamilton, Lei
Date Issued
August 20, 2026
Abstract
This paper outlines the SWaP-C (Size, Weight, and Power-Compute) considerations for implementing embedded agents for lightweight drone operations. It leverages a modern, open-source framework (TypeFly) to compare cuttingedge approaches in drone autonomy and document various performance tradeoffs between traditional and modern vision processing techniques. Research employs modern AI techniques for edge deployment of lightweight models for operation-focused workflows and benchmarks the performance of vision-capable Vision Language Models (VLM)s vs. traditional YOLO models for object detection on an open dataset of drone imagery. It explores various approaches of deployment and compares the implications of deployment regarding power, compute, latency, and performance.
Subjects
agentic systems
drones
edge AI
large language models
SWaP-C
sUAS
AI
MIT Department
Lincoln Laboratory
Persistent DSpace Link
