Started with overview of @Brian Kilberg's dissertation work.
Hardware goals are to use 5 SCuMs simultaneously.
Suggestion to write quick model of expected data to estimate performance in simulation before setting up experiment
Question: how will IMU data be combined? Averaging, weighting, etc.?
Could benefit from a hardware accelerator if successful. @Yichen Liu pitched a NN accelerator for robot control too. Current spec is 5 input, 3 hidden, and 2 output layers.