First On-Orbit Demonstration of a Geospatial Foundation Model

The NASA Astrobee has successfully demonstrated the first on-orbit application of reinforcement learning (RL) for autonomous control aboard the International Space Station (ISS). This achievement involved training a deep neural network using NVIDIA's Omniverse physics simulator, allowing the Astrobee to navigate effectively in microgravity environments. The results validate a new training pipeline that bridges the simulation-to-reality gap, showcasing the potential for RL in space robotics.

A new approach called Multi-Scale Visual Prompting (MSVP) has been introduced to enhance small-image classification tasks, utilizing lightweight, learnable parameters integrated into the input space. This method significantly improves performance across various convolutional neural networks (CNN) and Vision Transformer architectures while maintaining a minimal increase in parameters.

A new technique called ViT-SR has been introduced for Single Image Super-Resolution (SISR), utilizing a two-stage training strategy that incorporates self-supervised pretraining on a colorization task followed by adjustment for 4x super-resolution. This method simplifies residual learning by predicting high-frequency residual images added to initial bicubic interpolations, achieving notable performance metrics on the DIV2K benchmark dataset.

The US Naval Research Laboratory's APIARY experiment successfully demonstrated the first reinforcement learning control of a free-flying robot in space using the NASA Astrobee on the International Space Station on May 27, 2025. This experiment involved training a robust control policy in a simulated environment, enhancing the robot's ability to operate autonomously in zero-gravity conditions.

The ALDI++ framework has been adapted for security X-ray object detection, addressing the challenges posed by domain adaptation in real-world applications. This adaptation is crucial due to the significant variations in scanning devices and environmental conditions that can degrade model performance, as demonstrated through extensive experiments on the EDS dataset.