Brandon VanBuskirk

Kamigami Swarm Robots

• What: Lightweight underactuated swarm robots built on Kamigami chases and designed using end-to-end Reinforcement Learning, communicate on a ROS network to push or pull objects in their environment that would otherwise be impossible for a single robot to achieve.
• Why: Using end-to-end Reinforcement Learning methods like Model Predictive Control and Model-Based Meta-RL, these robots can move an object to a target location or through a desired trajectory by implicitly learning their own dynamics and the characteristics of their environment. In addition to this impressive performance, our sample efficient policy drastically cuts down on the training data required for a dynamics model of this type. The most comparable literature to our work, by A. Nagabandi et al, uses a dataset on the order of 10 thousand samples whereas we are able to achieve exceptional performance with as little as 10 samples
• How: Each robot is equipped with:
  1. RaspberryPi Zero - fast computation and network capabilities at a low mass
  2. Custom PCB - containing IMU and ADC for sensory input(currently not used, nor necessary)
  A peripheral camera aids the RL algorithm by tracking the location of the robots and updating destinations per unit via the ROS application, RVIZ.

Video and Image Transmission using Radio Waves

• What: A custom modem that compresses animated PNG files and using APRS packet format, transmits these packets to a reciever, which then is decoded and reconstructed.
• Why: Every day we use devices that seamlessly transmit and recieve information and very few people stop to question the magic at our fingertips. This project is an exploration into signal processing and communications, with lots experimental design along the way. Everything is coded from scratch(except Huffman Encoding - thank you MIT).
• How: Animated PNG files are transformed from RGB to Luma/Chroma channels and then processed separately. Using the 3D discrete cosine tranform, 3D laplacian quantization, and spiral, zero run-length, and Huffman encoding, the video is converted to byte code. This byte code is partitioned into APRS packet format and transmitted in NRZI byte stream. The reciever then uses a phase lock loop to detect the data stream and invert all encoding.

Compression: 240:1 (2.4MB to 10kB)

Takeaway: The goal of this wasn't to perfectly recreate the original video, but to find the limits of compression in which the recovered video is fairly close to the orginal but much much more efficient to transmit. This project taught me alot about communications and signal processing design. I had to learn many experimental video encoding techniques and made numerous experimental choices that ultimately did not improve my modellike PCA, frame differencing, motion detection, to name a few. But what's important is that I learned what works, and what doesn't, and how to take experimental concepts and apply them.

Directed Energy Lunar Rover

• What: In response to NASA's Big Idea challenge, this project was to propose and design a solution to the dilemma of traversing permanantly shaded regions(PSRs) of the lunar surface. We worked under alongside Professor Philip Lubin and adapted some of his work on directed energy as well as some of our own research to make it a viable solution.
• Why: The lunar surface is known to contain ice and could some day be used as a rocket refuel station for deep space missions, greatly preserving fuel on Earth-based launches. The problem is that the Moon's ice is mainly in PSRs(deep craters of the polar regions), meaning that solar powered rovers are no longer an option. Thus there is a big energy dilemma and few ways to successfully and affordably study the ice of the lunar surface.
• How: Our model is a system of two cooperative rovers. One with a laser and solar cells that remains at the rim of a crater collecting sunlight and directing energy to the recieving rover below, which deploys of its own for power supply.

Takeaway: This project posed many difficult challenges for our team to attempt to find solutions to. This project was a transformative experience for me as a Sophmore and taught me so much about insulation, power systems, optics, and design in general. We were ultimately eliminated from the challenge for having an overly ambitious design and righfully so. This was my first extracurricular research experience and showed me the beauty of space engineering and science for the first time.