skelarm

A lightweight, physics-based dynamics simulator for a configurable planar robot arm. skelarm treats the robot as a "skeleton" of N links and focuses on kinematics and dynamics — there is no collision detection or detailed shape rendering. The supported model is a horizontal, gravity-free plane: it models a planar arm lying flat.

What it does, in one pass: define a robot in TOML, pose it interactively (FK/IK), simulate its dynamics under torque control, run controlled tasks (reaching, curve tracing, trajectory tracking, MPC) from a single scenario file, and record every run to a self-contained log that replays, plots, exports to video, and — for headless runs — re-simulates reproducibly.

Where to go

• Getting Started — install and drive your first robot in five minutes.
• User Guides — task-oriented walkthroughs: configure a robot, pose and inspect it, simulate dynamics, run controlled scenarios, record and replay, teach trajectories, joint limits, and the tool & CLI reference.
• Developer Guides — use and extend the library from Python: API quick start, define a task, define a controller, architecture, and development & testing.
• Theory Reference — the mathematics behind the implementation, from forward kinematics to reaching control.
• API Reference — per-module API docs generated from the source (the Architecture overview lives under Developer Guides).

Capabilities at a glance

• Configurable robot — arbitrary planar chains from TOML: lengths, masses, inertias, centers of mass, joint limits, optional base offset.
• Kinematics — recursive FK for positions/velocities/accelerations, endpoint Jacobian, and numerical IK (Sugihara-style Levenberg-Marquardt and friends).
• Dynamics — Recursive Newton-Euler inverse dynamics, mass-matrix forward dynamics, fixed-step semi-implicit Euler and adaptive solve_ivp integration.
• Control — trajectory-tracking laws, human-like reaching controllers, and joint-space MPC, all configured from one scenario TOML and extensible at runtime.
• Recording & replay — self-contained *.sklog.npz logs with an analysis player, headless .mp4/.gif export, and reproducible headless re-simulation.
• Quality — fully typed, tested with pytest + hypothesis property tests, linted with ruff.