Rocket Academy - First Contact

Rocket Academy: FIRST CONTACT How do we prepare robots for places they cannot easily reach? FIRST CONTACT is a planetary robotics prototype that turns simulated missions into traceable, reusable experiences. Our vision brings together Earth, Moon and Mars terrain, AI-driven mission planning and robot simulation. Researchers, students and developers can explore scenarios, inspect robot decisions and build toward richer training datasets for autonomous exploration. The prototype includes planetary destinations, bounded rover plans, deterministic kinematic execution, mission replay and MCAP exports. Each recording captures time-stamped robot states, commands, coordinate transforms, planning provenance and outcomes, creating an inspectable record of a mission. MongoDB Atlas is designed as the mission data backbone.The backend links terrain sources, robot profiles, plans, runs and decision events. GridFS stores MCAP recordings with integrity hashes, connecting each episode to the inputs and versions that produced it. This provides a foundation for finding, comparing and governing simulation runs as the dataset grows. A Gemini adapter is designed to translate mission requests into structured, validated commands, while a local baseline planner can operate without model access. Atlas and Gemini integrations are implemented, but live-service verification is still pending. We’re building with scientific transparency in mind. Current episodes are synthetic kinematic simulations, not physically validated training data. Reference imagery is distinguished from measured terrain, and graphics-dependent globe interactions still need verification. Next, we aim to connect measured terrain with calibrated physics, simulated sensors and teams of autonomous robots. Semantic discovery and versioned datasets could help researchers find relevant experiences, evaluate behavior and improve future policies. FIRST CONTACT lays the groundwork for a new loop: plan a mission, simulate it, preserve the evidence, and learn from the result.

Team