FeFER

FeFER is an open‑source software library designed to provide advanced feature extraction and fusion capabilities for real‑time robotic perception systems. The project was initiated in 2018 by a team of researchers at the Institute for Intelligent Systems Research in Graz, Austria. Its core objective is to enable lightweight, high‑performance processing of multimodal sensor data, such as laser scans, depth maps, and inertial measurements, on embedded hardware.

The architecture of FeFER consists of a modular feature extraction engine, a data fusion optimizer, and a lightweight persistence layer. The extraction engine implements a suite of hand‑crafted descriptors (e.g., normal distributions transform, keypoint detectors based on point‑cloud curvature) alongside learned descriptors derived from convolutional neural networks. These are then fused by a Bayesian filter that reconciles the temporal coherence of successive sensor readings while accounting for sensor drift and occlusion.

FeFER is distributed under the MIT license and is compatible with the Robot Operating System (ROS) 2, providing a set of ROS 2 packages that expose the library’s functionality as services and topics. The most recent release, version 2.3.1 (March 2026), introduces a new GPU‑accelerated module for feature matching and adds native support for the Intel RealSense D435i camera.

The community around FeFER includes academic institutions, commercial robotics firms, and individual hobbyists. Contributions are managed through GitHub, and a collaboration portal hosts documentation, tutorials, and sample workflows. In addition to its use in autonomous navigation, the library has been adapted for applications in agricultural robotics, warehouse logistics, and augmented‑reality sensor fusion.