ESA GNC Conference Papers Repository

Dragos-Georgel Gogu, C?lin-George Mihalache, Florin-Adrian Stancu, Massimo Casasco, Olivier Dubois-Matra, David Merodio Codinachs, Fabio Vollaro, Julia Wajoras
Presented at:
Sopot 2023
Full paper:

As current and future space exploration missions are getting bolder, so do the requirements on the navigation algorithms and subsequently on their implementation in the avionics modules and on the associated data interfaces. To achieve the high level of autonomy and reliability required in today’s space missions, vision-based navigation making use of high resolution images and processed on-board at high frame-rates is mandatory. The increased complexity of these computer vision algorithms and the data fusion of measurements acquired from various on-board instruments and sensors mandate the development and use of high-performance avionics to provide one or two order of magnitude faster execution than today's conventional space-grade processors. As consequence, nowadays substantial efforts are done to develop high frequency, high accuracy navigations systems, to improve on-board spacecraft power consumption and to optimize the processing times of the computational loads to achieve near real time operation. This paper focuses on the development of GMVision, a highly versatile space-oriented image processing unit that can provide interfacing SpaceWire control and management of up to 8 links and the opportunity for integrating computationally demanding image processing algorithms. The GMVision board can provide various redundant Vision-based Navigation solutions for autonomous operations in space missions, as GMV has previously developed within different ESA activities (CAMPHORVNAV, NEOGNC-2, Lunar-Lander, PILOTB+, HERA, HIPNOS, ORCO, NEOSHIELD, HERACLES or Mars Sample Return) several image processing algorithms and navigation filters based on autonomous relative or absolute navigation, suitable for rendezvous-search-capture, active debris removal and in-orbit servicing, descent and landing into small or massive bodies or to accommodate computer-vision solutions for rover exploration. From the mechanical point of view, GMVision fits inside an envelope of 16 cm x 28 cm x 5.9 cm (dimensions adjustable by re-design), and it can be provided as independent boards with metallic enclosure box accumulating a total mass of approximately 1.5 kg, or it can be provided as rackable electronics boards to integrate with an OBC. The image processing board can adapt to different redundancy concepts, depending on mission needs or system requirements. The GMVision system provides 2 boards isolation for Image Processing Function and Interfaces function, as it is relying on a two FPGAs architecture for each board. Both FPGAs are rad-hard European BRAVE units with allocated external volatile and non-volatile memories: an NG-MEDIUM dedicated to interfaces control unit and monitoring, and a powerful NG-LARGE to perform as computer vision co-processor, improving the execution times with 2 orders of magnitude compared to space processors. The image processing algorithms functionality can be divided between the processing FPGA and a SW processor, making use of a HW/SW co-design approach, as GMVision can be interfaced via SpaceWire with an On-board computer. This is the case for the two use-case scenarios for the development of GMVision Mars Sample Return mission concept Vision-Based GNC Rendezvous with ERO platform or Lunar lander with both absolute and relative navigation. The SpaceWire interfaces allow TM/TC exchange between GMVision and up to 3 devices/instruments using nominal and redundant Spacewire links (nominal/redundant 1 on-board computer and 2 navigation cameras or eventually 8 different components). The design and development of the computer-vision algorithms for GMVision are facilitated by the architectural design of the processing FPGA code, which provides an internal interfacing wrapper to integrate the required image processing module satisfying a client-consumer simple interface. GMVision board can also include pre-processing functions for the navigation cameras or other sensors, as well as managing the redundancy concept. The high performance, fast and complex computer vision solutions are focused on the needed processing, so both FPGAs of the board are SRAM reprogrammable devices which allow flexibility and many options for the design and implementation of complex functionalities, such as high-data rate interfaces management and hardware accelerators. Thanks to the reprogramming capabilities it can also be possible to accommodate different computer-vision accelerators which are not used in the same moment of time by replacing partially or fully bitstreams in the processing FPGA to save a potentially needed second FPGA unit. The GMVsion concept is promising technology and it has high exploitation applicability also to other missions, such as Lunar Missions with Roadmap as a Smart sensor. Being reference architecture for complex high performance algorithms, the product is a feasible solution for space-based surveillance system, or to the Mars exploration missions. Interconnected to a space processor, GMVision is the perfect avionics platform for integrating vision-based system for autonomous rover navigation, implemented in full HW or as a HW/SW co-design with most computationally intensive operations implemented in FPGA, and those requiring additional precision implemented on a space processor. The TRL to be achieved by the GMVision concept is 6, so the product has not been qualified for flight yet, being built with commercial components. Nevertheless, extensive analysis were performed in terms of analysis and shock, and tests at ambient conditions. The roadmap towards an actual flight model is deemed with high probability of success, as all the electronics components of the existent EM have space qualified counterparts, so the flight model shall be able to withstand the harsh space environment. In summary, the paper presents a highly versatile, fast and complex avionics development deemed as co-processor for space computer-vision applications. The features, performance and the evolution of the concept from breadboard model to engineering model, as well as the roadmap towards a powerful flight hardware are presented.