Updates

The software division of the team has finalized the code for competition, while currently working on fixing any bugs. The largest component is the vision code for the detection of competition items. By using the YOLOv9 model, each pixel is evaluated to find patterns to identify objects. We will be manually fine-tuning it to better increase its accuracy. This new system is much better than our old method of detection which only used shape outlines and colors.

Mapping for our drone has also reached completion. Our vision code is fed by remotely capturing downloaded photos from an SIYI A8 mini camera, allowing Sparrow to chart its surroundings. Servo functionality has been implemented with no issues, and the camera code is nearly completed for both the simulator and the real drone. The team will continue to update the code with new optimizations as well fixing the few remaining bugs.

The past couple weeks were spent working on the payload delivery system. After brainstorming designs, it was between a winch design or parachute design. After testing, we could not get the winches to deliver the payload at a proper, slow speed. The parachute design fixed this flaw. We were able to repurpose an old design from previous years, with the addition of an updated attachment to hold a beacon.

The drop chamber had numerous variations, such as a vertical or horizontal rectangular prism. Electromagnets were considered, but required too much of a power draw. It was settled to use a horizontal rectangular prim with 4 bay-doors opened with servo-release mechanisms.

With the team’s completion of the SUAS drone build, a test flight was in order. After checking and tightening the props, fastening motor mounts, and safety checking the batteries and wires, it successfully completed a short hover test. Next, we did an 18 minute long flight with a lot of complex movement, testing different flight modes and getting a feel for how long a maximum length flight could last. We believe we are within the 30 minute threshold for competition. There were no glaring issues to immediately resolve after watching the flight, but we will still be continually fine tuning to improve speed and maneuverability!

We have been working very hard at getting the dronefully assembled, after finishing 3D printing its components. This includes getting our carbon fiber plates water jetted to shape, manufacturing our own batteries by spot welding individual cells together, and piecing together the wiring harness. We currently are working on finishing the payload mechanism.

We finally have an Unreal Engine environment that works with a simulated camera which brings us another step closer to testing vision models in the simulator. This means we have everything from the environment recreated for the competition: the field, a virtual drone that flies and works with our code, and a camera that is able to take photos of simulated models from the drone’s point of view. Our team has also finished all competition required code, which will be used on both the live and simulated models of the team’s drone.

The hardware team finished the carbon fiber layup of the drone’s main body, which will get water jetted in the upcoming weeks. We have also received all of our battery components and have started assembly. There are only a few more parts left to 3D print.

Similar strides have been made on the software side. A new docker container has been created for running simulation code to work on creating an environment to test vision models. We will be implementing the YOLOv9 model into our code to identify objects with machine learning. The mapping state has a first completion done on it and is now working on getting improved. New gimbal and camera code is now being implemented, along with old reformatted code to work with the addition of a mapping state.

After careful consideration, the team has landed on how the drone will be powered. 10 batteries, placed in pairs of two secured with a custom bracket each placed under its own arm to aid in balance. The last pair will be attached to the frame. The 48V batteries will be routed to a MAUCH Power Distribution Board, distributing 5V to the cube and telemetry radio​. The drone will also utilize a custom 37x41mm PCB.

Using Python to communicate with Ardupilot, the team had its first successful autonomous flight. This is a huge accomplishment!

The drone’s drop system is in the early stages of its design. Currently, we have a vertical caged structure with spools to hold some fishing line. The spool’s rod is controlled with the use of servos. Electromagnets were originally going to be used for their minimal power draw, but servos were chosen for their ease of installment and reliability.

Our new drone will be smaller than previous years’ designs, compacting its size and weighing less overall. The initial framing has already been designed, with slight alterations to come. The team has now begun looking into batteries and designing a harness to secure it.

The team has made the difficult decision to swap from PX4 firmware to Ardupilot. Our Flight Team has made great strides in mastering Dronekit to communicate with Ardupilot using Python code. The computer vision machine learning models are now fully set up, with the YOLOv9 module actively being trained to detect the wide range of objects that will be seen at SUAS. The mapping has been taken into a new direction, utilizing a coordinate-based algorithm to stitch images together compared to using recurring patterns and objects. Drone simulations are nearing completion for applying the drone’s code within the competition’s framework.