InFuse top level objective was to develop a data fusion framework offering a comprehensive set of data fusion techniques for perception and navigation, applicable to space robotics both in natural and man-made environments.
This overarching objective was fully achieved, with the open source release of the InFuse CDFF addressing both orbital and planetary applications, and both natural and artificial environment perception data handling. The set of data fusion capabilities implemented in the CDFF is quite comprehensive, allowing to address a number of relevant OOS and planetary robotics scenarios.
The first Specific Objective (SO1) was aiming to develop essential support functions for data fusion, and have them robust enough to fit in wide range of perception and navigation algorithms. A number of such fundamental data fusion algorithms was implemented (e.g. variations of Kalman Filter), and used in most perception and navigation algorithms released in the CDFF. Their robustness and performances were assessed only at DFPC level, not at individual (elementary) DFN algorithm level.
SO2 deals with perception related data fusion techniques. All the identified data fusion capabilities were integrated in the framework: target tracking and models reconstruction in particular. Alternative sensors data were also investigated, but more experimentally – related software was not integrated in the CDFF baseline release of InFuse.
SO3 addresses data fusion techniques for navigation. Localization in both structured and unstructured environments were tackled and covered in InFuse. For what concerns orbital applications, focus was on close to mid-range navigation.
In SO4, we proposed to develop a suite of tools supporting the access to data fusion techniques and products in the framework. 2 complementary data products management tools were implemented to handle data products in the frameworks.
SO5 dealt with the implementation of 2 reference implementations (RI) of InFuse: one orbital, and one planetary. The orbital RI was validated mainly with DLR OOS facilities and related data sets, while the planetary RI was validated with datasets obtained with the PEL facility of DLR, as well as with DFKI’s Sherpa platform (extended with a modular perception bench developed by DLR) and the Mana and Minnie rover platform of CNRS.
The dissemination of InFuse was good (6 peer reviewed papers). The consortium is now in the process of leveraging the results to publish journal papers (2 undergoing) and peer reviewed papers.
The InFuse CDFF was eventually released as an Open Source software on Gitlab, opening it to a wide community of roboticist – the framework is Space Robotics oriented and flavored, but definitely extends to a much larger community needs. We expect both space robotics and non-space robotics community to uptake InFuse’s results in various robotic applications. Gitlab link to the open source release:
https://github.com/H2020-InFuse(s’ouvre dans une nouvelle fenêtre)
