Get the latest tech news
Model Predictive Control in the Browser with WebAssembly
Commentary on software, robotics, and computer vision.
One advantage of this approach is that changes in physical system parameters (masses, drag coefficients, etc) are naturally accounted for by the dynamics model $f$, provided they can be adequately measured. I found that values in this range empirically produced a reasonable tradeoff between fewer optimization iterations and keeping the number of decision variables comparatively low. Typically when I find an issue, I save a log of the system state to JSON and then load this offline in a Python script (the optimization is also wrapped via nanobind) for further exploration.
Or read this on Hacker News
