Illinois AUV has limited resources and manpower for designing and implementing complex mechanical systems. Our school does not provide students with easy access to CNC machining tools, and few metal machining tools are available. With these limitations, our team focused on building a submarine that was feasible to construct with only a laser cutter.
Our team uses the NVidia Jetson TX2, a credit card-sized computer that has many hardware pins for connecting to various sensors without any modifications or extra peripherals. Our electronics fit completely in a 6"x12" tube, and we do not have to open the tube for any normal operation.
In our first year, we found that hand tuned classical computer vision algorithms were too sensitive to changes in light parameters. We are using the YOLO v3 architecture for detection, running on the Jetson TX2. To conserve computing power for other processes, YOLO is limited to run at only 4-5 fps, as the GPU is needed for other tasks, such as visual odometry.
Developing a physics-based simulator was one of the top goals for the team. We wanted to create a system where we can test our robot software stack in simulation to accelerate development and testing, even before a physical robot was complete.
|
|
|
|
|
|
|
|
|
