Welcome to Issue 20, where Transporter-18 carries orbital compute and onboard AI
The Transporter-18 rideshare that was a launch date last week is now on orbit, carrying both orbital AI compute hardware and an onboard-AI satellite on the same manifest.
On the Space4AI side, the hardware Issue 19 had on the calendar is up. Google's Project Suncatcher prototype, built with Planet, is in orbit with its Trillium accelerators, and TakeMe2Space flew an Nvidia compute satellite on the same rocket. Sophia Space and Redwire agreed to build the structures those orbital computers will need.
On the AI4Space side, that same rideshare carried the first commercial satellite built in Abu Dhabi, processing Earth-observation imagery onboard. Separately, researchers at the Daegu Gyeongbuk Institute of Science and Technology released an open simulator for testing the navigation methods a Mars rover would use.
Specifics below.
Did you know?
We have recently launched the Bits & Orbits Insights
Check out the growing set of independent dashboards, trackers, and maps on the
AI × Space frontier:
Space4AI
Transporter-18 launches orbital-compute payloads: Google Project Suncatcher M1 (Trillium TPUs) and TakeMe2Space MOI-1A (Nvidia)
On October 1, 2026, SpaceX's Transporter-18 rideshare carried orbital AI compute hardware to low Earth orbit. The flight had an unusual unifying theme for a rideshare: getting power to compute in orbit. We previewed the scheduled launch in Issue 19.
Google's Project Suncatcher M1 prototype, built with Planet Labs, rode Transporter-18 to orbit and put Trillium TPUs into LEO. The satellite is operating as expected. Google plans to collect weeks of in-orbit data on how those chips handle radiation and thermal extremes, and calls the flight the first step of Project Suncatcher.
Flying alongside it was MOI-1A from the Indian startup TakeMe2Space, which readied the satellite for the Oct 1 Falcon 9 rideshare. The bus carries Nvidia compute chips and serves 23 commercial, academic, and government customers. Those customers run ML pipelines for digital mapping, crop-yield prediction, land-use classification, and insurance risk assessment, all processed in orbit before downlink.
Sophia Space and Redwire sign MOU on orbital data center infrastructure
On September 30, 2026, Sophia Space and Redwire Corporation (NYSE: RDW) announced a Memorandum of Understanding to jointly develop and deploy scalable computing infrastructure in orbit. The agreement combines Sophia Space's modular orbital compute architecture with Redwire's technology in deployable space systems, solar power generation and distribution, heat management and rejection, spacecraft integration, avionics, and systems engineering.
Under the agreement, the companies will explore collaboration across near-term demonstration missions and longer-term orbital data center architectures. The technical core is Sophia Space's TILE architecture. Its modular TILE system integrates high-performance computing, solar power, and passive radiative thermal management into scalable systems designed for AI inference, data processing, and other compute-intensive workloads in orbit. The companies frame the partnership as addressing a central challenge in the emerging orbital computing market: moving from individual technology demonstrations toward the repeatable, scalable infrastructure needed for a global build-out of orbital data centers.
The two companies intend to jointly pursue select U.S. government and commercial opportunities, including potential programs involving DARPA, the Air Force Research Laboratory, the Space Development Agency and the National Reconnaissance Office, across concept development, joint proposals, internal research and development, customer engagement, and international markets. Redwire, which employs approximately 1,400 people across North America and Europe, contributes the space infrastructure and engineering base to Sophia Space's TILE compute payloads.
AI4Space
Orbitworks launches Altair-1, first UAE-built commercial satellite with onboard AI for multi-sensor EO
On 1 October 2026, Orbitworks launched Altair-1, the first commercial satellite built in the UAE, aboard SpaceX's Transporter-18 rideshare mission from Vandenberg Space Force Base in California. The satellite was integrated by Loft Orbital, acquired signal after deployment, and entered its commissioning phase. It is the first of a planned 10-satellite AI-enabled Earth observation constellation.
Altair-1 was designed, assembled, integrated and tested at Orbitworks' facility in Abu Dhabi before being shipped to California in August. Orbitworks is a joint venture between Abu Dhabi's Marlan Space and Loft Orbital. Each Altair satellite carries five integrated sensors covering sub-meter optical, shortwave infrared, thermal, hyperspectral and radio-frequency data. Onboard NVIDIA GPUs process that data into analysis-ready intelligence within minutes, cutting reliance on ground-based processing.
The remaining nine satellites are scheduled to launch through 2027, and once the constellation is fully operational it is designed to revisit any point on Earth roughly every four hours. Target applications include disaster response, maritime awareness, environmental monitoring, food security, infrastructure surveillance and national security.
Dr Ahmad Belhoul Al Falasi, Minister of Sports and Chairman of the UAE Space Agency, congratulated the launch and tied the milestone to the National Space Strategy 2031.
Researchers release MarsLab, an open-source Isaac Sim-based Mars rover simulator for autonomous navigation benchmarking
On September 28, 2026, Hoyun Kim, Beomsu Kim, and Giseop Kim, of the Department of Robotics and Mechatronics Engineering at the Daegu Gyeongbuk Institute of Science and Technology, released MarsLab, an open-source simulator for Mars-rover navigation. The work was accepted for publication at the 2026 International Conference on Space Robotics (iSpaRo). MarsLab does not fly a new model. It gives navigation methods a shared Martian testbed: simultaneous localization and mapping, and visual place recognition under changing dust and light.
MarsLab runs a Perseverance-class rover model in NVIDIA Isaac Sim (version 5.1, with ROS2 Jazzy) and combines HiRISE-derived and procedural terrain with configurable rock, crater, solar-illumination, and atmospheric-dust settings. The runtime publishes RGB, depth, RGB-D point clouds, LiDAR, IMU, wheel odometry, and ground-truth pose over standard ROS2 topics, giving each algorithm a shared ground-truth trajectory for head-to-head comparison. The authors benchmark simultaneous localization and mapping across sensing modalities, dust levels, scene geometry, and route length over three courses: Mars Base (440 m), Main Crater (393 m), and Grand Canyon (648 m).
The mapping results quantify how vision degrades where LiDAR holds. LiDAR-based mapping stayed sub-metre across all scenes at 0.13–0.38 m absolute trajectory error RMSE, while ORB-SLAM's RGB accuracy degraded 7.9× under dense dust on Mars Base, and monocular tracking lost the route entirely in Grand Canyon. The LiDAR method (MOLA) was unaffected because the dust model is radiance-only.
For visual place recognition, repeated Mars Base traversals tested whether a rover can match a camera frame to a place it has already seen, under illumination and dust changes. BoQ ranked first in both protocols, reaching 91.44% Recall@1 across a day-to-dark illumination change and 94.65% under dust, with NetVLAD following and AnyLoc recovering by Recall@10.
Till next time,
Meta-beat Column of this week
Read also about the AI Pipeline that sits at the core, producing this Newsletter, including its ups and downs of this week: