I realized that when I was doing my setup last year, my current readings were way too high, leading me to prematurely failsafe with a nearly full battery. I did a manual adjustment with a dc current clamp and now tried another flight, which completed fully with over half the battery remaining. The whole flight wasn’t logged, though (despite my clearing the logs beforehand) and a second, empty log file was also present.
Any ideas why this could have stopped logging? The flight continued like normal.
I might* get a new SD card tomorrow and try to do another set of magfit and quiktune flights tomorrow before it rains. I really want to get onto the sprayer part of it.
Which of these settings should I “trim” down? Especially if I want to run the current SD card?
While I remain very critical of LLM use in general, I am a proponent of its proper use. I’ve seen some promising results lately with LLM log analysis, though I think one needs to be familiar with such analysis to begin with before turning the “robots” loose. At any rate, no one wants to read all that generated drivel, and I think the point has been made here.
@OmkarSarkar204 is integrating heuristic-based log quality analysis that does parameter change recomendations inside AMC. The heuristics results are presented to the user and a LLM will be fed with them.
So even unexperienced users with no engineering background should get the most out of it.
AMC+LLM and everyone will get correctly working vehicles with as little effort and time as possible
That’s awesome! I’ll follow his profile to watch the project. It seems like he’s using a heuristic approach to determine the params and mostly using the LLM to interpret/explain the reasoning to users in natural language.
I definitely think this is a wiser approach (at least for now) than giving the LLM free-rein over the actual configuration.
Even during my log analysis, the latest Claude Model gave me what I consider to be some poor advice; it suggested that I increase the log buffer size which was unnecessary, would reduce free memory, and would make me use an out-of-range value. (400 instead of 200 which is the max) This was one of the two “immediate” changes it confidently suggested, but it immediately reversed course and withdrew the suggestion when I had Grok double-check whether this setting change was a good idea.
When ChatGPT first came out, I tried to use it for adding some functionality to a simple C/C++ firmware which ran the irrigation controller for my garden. It would just “break” the code and couldn’t even blink an LED. Now, Claude vibe-coded a Home Assistant-compatible version for me on the first attempt which “just works out of the box”. (in YAML)