Good morning. It's Thursday, June 18th, 2026.

**Story one.** NASA has picked Relativity Space to build a Mars orbiter mission called Aeolus, launching in 2028. This is the first time NASA has handed a private company — one whose rocket has never actually reached orbit — the job of delivering science instruments to another planet. Relativity will design and build the spacecraft to carry four instruments that will measure Mars' atmosphere: daily global dust profiles, wind patterns, and temperature data. It also doubles as a communications relay for whatever's operating on the Martian surface. The twist: Relativity was acquired last year by former Google CEO Eric Schmidt, who installed himself as CEO and is betting the entire company can deliver this on a tight timeline — before SpaceX, which ironically has never sent a science mission to Mars either. Some analysts are calling it a major risk. NASA's public-private partnership model is stretching budgets, but Relativity's Terran R rocket hasn't reached orbit yet. If it works, it's huge. If it doesn't… well, past NASA startup partners have gone bankrupt.

**Story two.** Stellantis, Wayve, and Uber announced a three-way partnership yesterday to develop and deploy Level 4 driverless robotaxis globally. Stellantis handles the vehicles — they're building dedicated L4 platforms with embedded sensors designed from the ground up for autonomous operation. Wayve provides the AI driving stack, using their end-to-end approach that doesn't require city-by-city mapping. And Uber distributes through their global mobility network. This builds on existing relationships: Uber and Wayve are already deploying robotaxis in London and Tokyo, and Stellantis and Wayve just signed an L2-plus deal in May. This matters because it signals industry convergence toward the vehicle-plus-AI-plus-network model, where no single company tries to do everything. Compare this to Waymo's vertically integrated approach. The Stellantis-Wayve-Uber model could scale much faster if the AI is truly as adaptable as Wayve claims — no remapping per city.

**Story three.** At CVPR this week, NVIDIA announced ENPIRE — a closed-loop framework where AI coding agents actually run robotics research on real hardware. The system was built by NVIDIA's GEAR Lab with Carnegie Mellon and UC Berkeley, and it's open source. ENPIRE can reset physical scenes, run experiments on actual robot hardware, observe results, and iterate — all autonomously. This is robotics research accelerating itself. The other big story from CVPR: NVIDIA's physical AI agent skills, helping developers speed development of autonomous vehicles and robots. Combine this with NVIDIA's Isaac Root humanoid reference platform and the 50,000 humanoid robot shipments projected by end of 2026, and the pattern is clear: NVIDIA is trying to own the entire physical AI stack, from simulation to training to deployment. The ChatGPT moment for robotics, as Jensen Huang called it, is happening. And AI agents aren't just writing code anymore — they're running experiments on physical machines.

That's all for today.