Zebrafish Embryo Pick and Plate

  For some quick context on why such a machine was even needed, the Sinnhuber Aquatic Research Lab performs toxicology research using Zebrafish. This means tha the lab is also a breeding facility, generating a few thousand eggs a day, which need to be processed and isolated prior to being introduced to experimental conditions. At one point in time, these processes were done by hand, with all researchers spending a significant portion of their day simply prepping the embryos. Later on, they hired a contractor to design an automated solution to the isolation problem, and while they delivered, the units were industrial overkill, using massive SCARA arms and enclosures to pick up embryos which were roughly half a millimeter in diameter. They were also very expensive, costing around $150k per machine, and the lab had four installed. Thus, the engineering team was tasked with cost and size reducing them so that more could fit in the same space and throughput could be higher.

  My coworker, Dylan Thrush, and I got to work. He focussed on the mechanical side, while I worked on electrical and software. I had recently been learning PCB design after becoming more familiar with it through the OSU robotics club, and decided that a good place to start would be to create a motion controller. Dylan and I had already landed on a simple stepper-motor based system to more than meet the needs of a task like this, and both of us already had experience with stepper-based CNC machines, making it a great jumpoff point. My first PCB was unfortunately a massive failure, because while I did include a quad stepper motor driver, microcontroller, general purpose I/O, and usb to serial interface, I failed to export the gerbers correctly. This resulted in a PCB with no drills, making it completely useless (outside of a good learning experience)! It was also just a poor layout overall, which isn't surprising considering it was my first ever PCB design. If you want to see this embarrassing result, check out the PCB section at the end of this page!

  For the second revision, we'd made some progress in other places, but most importantly had decided that we would use a Beaglebone Black single-board-computer to run the whole system. In an effort to simplify the assembly, I decided that the next revision would include headers to directly mount the Beaglebone to the unit, providing power and a serial interface over those pins. This version actually worked as expected, with only a few very minor bodges to the microcontroller's crystal, and some bulk capacitance on the power input. Since things were working, I began writing embedded C to control motion through some higher-level interfaces. Not long into this process, I realized I might have bitten off more than I could chew. Not only was I having to learn the deep ins and outs of microcontroller programming, kinematics, and serial interfaces, but I would still have to greatly improve my Python skills, learn to create graphical user interfaces, and figure out how to detect embryos using a camera and computer vision. While I was sad to scrap this, we decided to fall back on a motion controller called TinyG, which would simply require gcode to be sent over serial to function.

  Dylan built up a mechanical testbed, which we could mount a camera to, and I began to focus on embryo detection. I was relatively new to Python, but quickly found the Qt framework for building decent-looking interfaces, and OpenCV for providing generic detection capabilities via webcams. Over the next few years (remember, we were students doing this part time, and working on other projects simultaneously), I eventually created a fairly comprehensive user interface that would allow researchers to tune detection and motion parameters on-the-fly. This was shown on a touchscreen that the beaglebone plugged into, making it quite intuitive. Dylan had also created a mechanical foundation providing repeatable alignment for a petri dish with embryos, the 96-well plate for placement, and a waste container to get rid of extra water in our metal pipette tip that would be used to pick up the embryos. I'd also created some very bright lighting boards to mount to the unit's extrusion, illuminating the embryos from the side, something that was absolutely required for consistent detection with the camera. After a long journey, we finally delivered multiple units to the lab! Seeing such small devices performing the same task next to the behemoths which were the prior versions was quite amusing. You could easily fit a two by two grid of these in the working area of each of those machines. Overall, this was a highly ambitious project, but it developed some of my most successful skills I gained while at OSU while accelerating the research at the lab!