Peter Deutsh, Muchen He, Arthur Hsueh, Wilson Wang, Ardell Wilson
The top priority of the three objectives are:
Integration of electronics with the drone is the most important because we need the understand how much we can do on a drone – exploration rather than a specific package
Video transmission because it is something that’s needed
Machine learning model
Ultimate plan: directly compiling machine learning model into a FPGA hardware. We are not worried about the machine learning model, but we’re more concerned about a compute platform that has programmable hardware that is compact and light.
We need the biggest FPGA we can find that can be reprogrammed for future purposes.
Updated goal for flight time to 5-10 minutes.
We need an easy way to reprogram the FPGA while the device is on board.
Quality goals:
Durability depends on the drone
(Repeat) flight time is at least 50% of the un-attached flight time
Accuracy of ML model is not as important
Performance - real time processing and transmission is preferred
Accuracy - the intersection of the bounding box with the image
Todo:
make a budget and selection spreadsheet on potential FPGAs.
Find the most complex and riskiest part of the project and try to tackle that first. In my (Muchen’s) opinion, given that the system integration is the top-priority requirement, I would start getting a DE1-SoC with some computationally heavy design running and computing the power requirements. Having a WiFi modem onboard and interfacing with them is also important.
Ask ICICIS for a drone (for research) - loan it to Mieszko Lis. If none is available, propose a spreadsheet of product selection of FPGA + drone combos.