PyGeometry is a Python package that implements common operations on the differentiable manifolds usually encountered in computer vision and robotics. Implemented manifolds: R^n, S^n, SO(n), SE(n), T(n). Implemented operations: conversion between representations, geodesic distances, interpolation, random sampling. The design goal is to have a set of well-tested primitives: I’ve been burned too many times from having used buggy functions. PyGeometry is paranoid on program correctness. It uses PyContracts to validate input and return values. stochastic_testing (another experimental library) is used to check the correctness of the random sampling operations.