Lunasa Ltd. (Lúnasa) is a UK-headquartered SpaceTech startup founded on the 1st of March 2021.  The company is revolutionising space exploration by empowering intelligence into spacecraft navigation systems. Its first product is an Autonomous Rendezvous Proximity Operations (RPO) technology which enables the execution of complex missions from satellite refuelling and active debris removal to cargo transportation to space stations and deep space planetary exploration. Lúnasa has been backed by leading deep-tech private investors and has achieved several industry awards from leading organisations such as the UK Space Agency and the European Space Agency. With a Team of 8 and growing, Lúnasa has built expertise in AI/ML, satellite navigation, space systems, and much more. 

Project Summary:

3D reconstruction is the ability to deduce the shape and appearance of an object from regular images. In a space context, understanding the shape of an object can be very useful for reasoning about the purpose of a spacecraft, eg. friend of foe? space telescope or space debris? An accurate 3D model also enables pose estimation, facilitating autonomous approach and docking. The goal of this project is to develop a reconstruction pipeline for in-orbit accurate 3D representations. 

Project Description:

The objective of this project is to create a reconstruction pipeline with the capability to generate precise 3D representations under the constraints imposed by space-related conditions. These constraints primarily encompass various types of imagery, including:

• In-space imagery, which is a challenging acquisition due to its limited availability in the public domain. 

• Simulated in-space imagery, 

• Realistic rendered imagery, generated through an in-house rendering pipeline. 

• Option to collect original imagery as needed. 

A major requirement for this project is that the developed pipeline must be compatible with edge-based computing. To facilitate this, we provide an Nvidia Jetson Orin AGX. The emphasis on edge computing ensures the practical applicability and efficiency of the developed pipeline in real-world scenarios. As a result, the project aims not only to achieve accurate 3D representations but also to do so in a manner that is operationally viable within the unique challenges posed by space-related imaging conditions

Applicant Specification:

We are looking for an ambitious, high-calibre, and motivated candidate to work on innovative computer vision technology for space proximity operations and docking applications. The intern must be comfortable with working in a team under agile development methods, interacting with researchers and engineers, and able to perform high-quality research in collaboration with the other team members. 

Minimum requirements:

• Understanding of the math and principles of machine learning algorithms 

• Knowledge of Python and PyTorch (or TensorFlow but willingness to work on PyTorch) 

• Knowledge of how to implement, train and optimise neural networks. 

• Ability to work in a team and individually

Desirable requirements: 

• Experience implementing computer vision algorithms from theoretical concepts to the actual application (from scratch or based on an existing high-level framework). 

• Exposure to the engineering problems involved in deploying machine learning algorithms to hardware with limited resources. 

• Knowledge of C and/or C++ 

• Experience with OpenCV and exposure to new 3D reconstruction techniques (NeRF, Gaussian splatting)

Process details:

• 12 weeks minimum fixed term contract to be agreed with the successful candidate

• In-person Induction Event is to be held, and an in-person/hybrid Showcase Event

• 5 days' holiday to be taken during the placement