New features of pytransform3d for robotics research: transformation graphs, uncertainty, and 3D visualization in Python

Rigid transformations in three dimensions (3D) are fundamental tools in robotics, computer vision and graphics, and physics simulation. However, there are many pitfalls for practitioners: various alternative representations of rigid transformations and uncertainty, each with its own competing conventions, sometimes without clear documentation, ambiguities in representations, singularities, and difficulties in handling many transformations in complex systems. This article presents the new features of pytransform3d, a Python library for working with rigid transformations in 3D that helps tackle these difficulties. Specifically, we show applications of its new batch data processing functions, uncertainty propagation, and 3D visualization. Furthermore, we present applications of its transformation graph TransformManager, a tool inspired by tf of the Robot Operating System (ROS), which has a subclass for loading robot kinematics and a new subclass that handles time-varying transformations. In examples from imitation learning, kinematics, state estimation, and log data processing, we demonstrate these features and show that only a few lines of code are sufficient to solve these complex problems correctly. The explicit terminology of the library makes this code readable and reviewable. The library has 100% test coverage of its core features, has a broader feature set than comparable libraries, and handles batch processing of spatial data in fewer lines of code and faster than its alternatives.