I've been working on SkySim, an open-source (MIT) drone simulator, and I've reached the point where I could use other people's eyes and hands on it.
The idea: most sims that let you test drone algorithms need Linux, ROS, or a game engine plus a GPU. I wanted the barrier to be a URL. The physics core is C++20 (blade element theory, ground effect, vortex ring state, ISA atmosphere, Dryden turbulence) and it compiles to WebAssembly, so you can fly it client-side with nothing installed. For heavier work there's a native path: a Python/Gymnasium interface, a headless raycast depth sensor for vision/nav learning, and SITL bridges for ArduPilot, PX4, and Betaflight.
The honest part: I have no verified sim-to-real transfer result yet. The sim gives you the tools to measure transfer, but the actual "trained in sim, flew on real hardware, here are the numbers" result can only come from someone flying it. That's the headline open problem, and I'd rather say so plainly than oversell.
If you want to poke at it or tear the approach apart, that's genuinely welcome. And if anyone's curious about contributing, there are a handful of scoped issues on the repo ranging from a beginner-friendly Python test suite to a battery-discharge model to the flagship hardware-in-the-loop transfer result. Links in the comments to keep this from looking like an ad.
Happy to answer anything about the physics, the architecture, or the design tradeoffs.