VOLTA SPACE VISION
Volta experience in Space starts with the tendering for European Space Agency for the AO/1-10356/20/F/CP: Artificial Intelligence for Terrain Relative Navigation in Unknown Environment (ATENA). Volta team demostrated in (ESA words) “Excellent experience in commercial applications of AI-based processing; very good undestanding of the study objectives; the bidder shows the capability to generate very accurate images representative of the reference scenario; The contractor performed a first iteration of the development loop of LIKA-Net applied to the use case, which give confidence in the feasability of the project; accordingly under this criterion the contractor has been considered between good and very good.”
This is LIKA-Net is an end-to-end collection of neural networks for visually estimating the distance, pose and other features relative to an unknown celestial body (= seen for the first time, no accurate model available).
The above video is 5280 times faster than realtime, it is shoot along the Z-shape trajectory relative to an unknown asteroid. Said computer-generated asteroid has an average diameter 780 meters, with rotation period 2.26 hours and with a moon of 170 meters; the orbital period of the moon is 12 hours. The FOV of the camera is 5 degrees, and the minimum flyby distance is 1 Km from the main body (minute 4:37 of the video)
LIKA-Net prediction is shown in the top-right corner, while the ground truth is in the top-left corner. In the bottom left corner there is the movement of the spacecraft along the Z-shape trajectory (X axis perpendicular to the plane).
Generative Adversarial Networks (GANs) are a family of deep learning algorithms used for image generation. Many applications have been explored in the literature, including content enhancement and partial modification of images. In the context of LIKA-NET, GANs are used to process the image of the asteroid in order to reconstruct the dark side. This can be useful for various applications:
● Compute a coefficient given by the ratio between the visible pixels and the reconstructed ones. This quantity indicates the visible portion of the body and can be used by the onboard software to autonomously decide when to acquire images
● Integrate a traditional centroiding algorithm and improve the accuracy (centroiding works under the assumption of globular shape, while our GAN-approach is perfect with irregularly shaped asteroids)
VoAG (Volta Asteroid Generation) is the procedural 3D modelling tool that runs on Volta’s server GPUs. With VoAG, the surface generation procedure that can model the asteroid surface with several features:
● Varying amounts and sizes of boulders of various materials. We distinguish C- (clay and silicate), S- (silicate and nickel iron) and M- (nickel iron) asteroids and we consider the different materials properties when modeling them.
● Craters and other irregularities. The VoAG software can procedurally modify the surface of an asteroid and produce craters, hollows and other irregularities at different scale, shape and density.
● Rubble pile asteroids generation, starting from a simulation based on the “collisional evolution theory”. Although computationally (very) heavy, this procedure allows to obtain a high level of realism and details.
Contact us for evaluating and getting a license of our LIKA-Net software: email@example.com
Proximo™ is the «general-purpose service robot» by Volta. Proximo is fully designed around A.I. for serving the mass market in everyday life as well as in space. Proximo will be presented in public in Q2 2024.