The Same Chip Running Our Survey Robot Is Going to Space
- Patrick Duggan
- Mar 26
- 5 min read
Updated: Apr 25
# The Same Chip Running Our Survey Robot Is Going to Space
*March 26, 2026 — DugganUSA*
Last weekend I was on my hands and knees in a house in Connecticut, calibrating a LiDAR by pointing a robot at a wall and reading the angles. The robot runs YOLO v8 on a Jetson Orin Nano. It detects people, greets them with a privacy disclaimer, detects cats, says "pss pss pss," and drives itself around a building scanning BLE devices and posting telemetry to InfluxDB.
This week, NVIDIA announced the Space-1 Vera Rubin Module — data-center-class AI for orbital platforms. The edge computing tier? Jetson Orin. The same chip. The same architecture. The one I was debugging on a desk next to a LiPo battery and a USB cable.
The chip that says "pss pss pss" to cats is going to orbit.
What's Already Flying
More AI is deployed in space right now than most people realize:
**CogniSAT-6** — NASA/JPL's first truly autonomous satellite mission, July 2025. Onboard AI decides where to point instruments and what to observe. No ground control in the loop. The satellite sees something interesting, evaluates it, and acts. In under 90 seconds.
**ESA Hera** — Launched October 2024, currently navigating autonomously to asteroid Dimorphos using the same principles that underpin self-driving cars. Multi-sensor fusion, environmental modeling, independent decision-making. Flying right now.
**Φsat-2** — ESA's edge computing AI satellite, launched August 2024. Onboard machine learning for Earth observation. The satellite processes imagery locally instead of downlinking everything to ground stations.
**Stanford ML on ISS** — First machine learning deployed to robots aboard the International Space Station in 2025. Robot motion planning improved 50-60% over ground-computed trajectories.
**Tianwen-1** — China's Mars orbiter autonomously targeted and imaged 3I/ATLAS, the third known interstellar object, in March 2026. Autonomous observation of an object that won't come back.
What's Coming
**Rosalind Franklin rover** — Autonomous Martian navigation and science target selection. The rover decides what's worth investigating without waiting 20 minutes for a human to answer.
**Lunar Gateway** — AI operations where communications delay makes human control impractical. The station needs to make decisions between contact windows.
**SpaceX Starship** — AI-based guidance, heat shield diagnostics, autonomous orbital adjustment and landing. The vehicle that's supposed to go to Mars needs to think for itself.
**ESA/JAXA MMX** — AI-powered landing on Phobos. Autonomous sample collection. Return to Earth. The moon's environment is unpredictable enough that pre-programmed sequences won't work. The AI has to adapt in real time.
The Latency Problem Is the Whole Point
Mars is 4-24 minutes away by light. The Moon is 1.3 seconds. Even LEO has enough latency that real-time human control of complex operations is impractical.
The answer is the same answer we arrived at with the survey robot: the AI has to be onboard. It has to perceive, decide, and act locally. You can't remote-control a rover on Mars any more than you can remote-control a robot in a hallway when the WiFi drops.
Our robot's autonomy loop: YOLO detects objects → GPT-4o Vision describes the scene → motor controller plans a path → BLE scanner logs devices → voice system greets or narrates → all of this runs in a 120-second survey cycle while the robot drives itself around avoiding furniture.
A Mars rover's autonomy loop: cameras detect terrain → onboard classifier identifies science targets → navigation system plans a safe path → instruments collect data → all of this runs while the rover drives itself around avoiding rocks.
Same architecture. Same constraints. Different gravity.
The Hardware Convergence
The $2.36 billion space AI market (2025) is projected to reach $15 billion by 2034. NVIDIA is putting Jetson Orin in space-rated configurations. The same chip running inference on our robot at a desk in Connecticut is being radiation-hardened for orbital deployment.
Microsoft pulls this feed daily. AT&T pulls this feed daily. Starlink pulls this feed daily. Get the DugganUSA STIX feed — $9/mo →
This isn't a metaphor. It's the same silicon. The neural network weights that detect a person in a hallway and trigger a privacy-first greeting are architecturally identical to the weights that detect a science target on a Martian surface and trigger an autonomous observation.
The difference between our survey robot and a Mars rover isn't the AI. It's the launch vehicle.
What We Learned That Applies
Building an autonomous robot that drives through a house taught us things that translate directly to space autonomy:
**Sensor fusion is mandatory.** Our LiDAR had a vendor SDK with a checksum bug. We wrote a custom driver. The factory camera system was already running, so we integrated with it instead of replacing it. Space missions face the same reality — sensors misbehave, drivers have bugs, and you integrate with what's available because you can't FedEx a replacement to orbit.
**Graceful degradation matters.** When our AI node can't reach the cloud API, it falls back to offline mode. When the camera feed drops, the robot keeps driving on LiDAR alone. When voice synthesis fails, it stays quiet instead of crashing. In space, graceful degradation isn't a nice-to-have — it's the difference between a $500 million mission and a $500 million debris field.
**The cat detection protocol.** We built a rule: detect cat → stop all movement → say "pss pss pss" → hold position. That's not a joke — it's a safety interlock triggered by object classification. Mars rovers have the same pattern: detect unexpected obstacle → stop → evaluate → decide. The cat is the rock. The protocol is identical.
**Stop-steer-creep.** Our robot's plastic wheels skip on hardwood if you steer under power. So we implemented stop-steer-creep: stop, turn, then creep forward. A Mars rover on loose regolith has the same problem. You don't turn under power on an uncertain surface. The physics don't care whether the surface is oak flooring or Martian dust.
The Point
The space AI market is a $15 billion opportunity built on the same technology stack we deployed in a house in Connecticut. Jetson Orin. YOLO. LiDAR. Autonomous navigation. Sensor fusion. Graceful degradation. Edge inference where latency makes cloud computing impossible.
We built a survey robot that greets people, avoids furniture, scans wireless devices, and says "pss pss pss" to cats. The same engineering — the same chip, the same architecture, the same design patterns — is going to navigate asteroids, land on moons, and drive across Mars.
Space is just a building with worse WiFi and no cats.
*Patrick Duggan is the founder of DugganUSA LLC. The survey robot runs ROS2 on a Jetson Orin Nano with custom LiDAR, YOLO v8, GPT-4o Vision, and a BLE scanner. The same Orin architecture that NVIDIA is deploying to orbit. The STIX feed is free at analytics.dugganusa.com/stix. The cat detection protocol is not yet rated for space.*
*Her name was Renee Nicole Good.*
*His name was Alex Jeffery Pretti.*
The cheapest, fastest, most accurate threat feed on the internet.
275+ enterprises pulling daily. 1M+ IOCs. 17.4M indexed documents. We beat Zscaler by 43 days on NrodeCodeRAT. Starter tier $9/mo — less than any competitor’s sales demo.
Comments