multicalc does exact calculus: derivatives, integrals, linear algebra, ODEs and spatial math, with zero heap, no panics, no unsafe. The same code runs from a 64-bit server down to a bare-metal chip.

Two big improvements: Every PR is now tested on six targets (x86_64, aarch64, and four bare-metal chips under QEMU) with hard limits on size, stack, and accuracy. And every number is checked against external libraries like numpy and SciPy down to the last few bits, so your microcontroller gets the same answer as your workstation.

What else is new in 0.8.0:

• Spatial math: quaternions and Lie groups (SO2/SE2/SO3/SE3) for rotations and rigid-body motion.

• ODE solvers: fixed-step RK4 and adaptive RK45.

• f32 support: the whole library (autodiff included) now runs in single precision too.

• Cleaner errors: typed, per-module error types.

• More tooling: a new RISC-V target, vector!/matrix! macros, and auto-generated accuracy tables checked in CI.

Where it's headed: This is the start of a fully integrated no_std robotics stack — one codebase from a tiny MCU to a 1 kHz control loop. Next up: Kalman filters and controllers (PID, LQR), then three live robot demos — a Roomba-style base, a quadrotor, and a Franka Panda arm. The arm IK demos are already running.

Come build it! The backlog is carved into small, well-scoped issues, with the easy ones tagged good first issue. If you like numerical methods, autodiff, embedded Rust, or robotics, I'd love the help.

• crates.io: crates.io: Rust Package Registry

• Repo: GitHub - kmolan/multicalc-rust: Calculus, autodiff, nonlinear least squares & linear algebra in safe, no-std rust. Works on any target: 64-bit servers to bare-metal microcontrollers · GitHub

• Good first issues: Issues · kmolan/multicalc-rust · GitHub

1 post - 1 participant

Read full topic