Skip to content

Latest commit

 

History

History
42 lines (31 loc) · 2.51 KB

PROPOSAL.md

File metadata and controls

42 lines (31 loc) · 2.51 KB

Dragonfly Missile Command: Neuromorphic Systems for Interception and Flight

Abstract

Our project aims to simulate the structure of the dragonfly brain to increase drones' efficiency in detecting missile trajectories. We take inspiration from the neural structure of the dragonfly brain to create the neural network architecture with a fully-connected layer, and convolutional layers. As a final step, we plan to implement our architecture on Akida neuromorphic chips to showcase how edge AI can enhance performance.

Objective

To design and simulate a basic biomimetic drone system using dragonfly-inspired neural and flight dynamics, with a focus on building a prototype capable of basic interception tasks and showcasing energy-efficient operation.

Design methodology

The first focus would be on the mechanism of our approach- To simplify these principles(dragonfly inspired) to a more functional neural architecture and possibly also the flight dynamics model. The second would be the architecture- to have a neural model and a flight model. Neural would be using SNN/CNN and flight would be creating or simulating a flapping wing dynamic in software based on the principles. After all this, we can think about integrating Akida which we can do using online tools.

Evaluation Metrics

Energy efficiency, Accuracy of the Simulation, and its feasibility

Hopefully(for the next 25 days), 1. basic knowledge of the required tools, 2. Development and testing, 3. Documentation

Steps

Getting started Finalizing proposal (11/19 at latest) Boilerplate akida sim (?) Version control (git and overleaf) (?)

Development

Map paper’s method for matrix translation -> implement as ANN + CNN Ideally, convert to SNN for maximum hardware compatibility Run the simulation, then on-chip Workshop : ) Presentation Planning an outline (?) Writing report

References/Sources