RocketSmith: AI agent designs and builds high-powered rockets with 84% accuracy

RocketSmith, developed by Peter Pak and five colleagues, is an agentic system that revolutionizes high-powered rocket development by automating design, manufacturing, and optimization. The system integrates intelligent automation of software tools to validate flight stability and generate parametric design components for assembly. It employs a collection of subagents and skills that enable optimization workflows via iterative zero-shot or human-in-the-loop processes. The team used RocketSmith to create four distinct high-power rockets with varying motor and assembly configurations, all leveraging additive manufacturing (FDM 3D printing) for fabrication. Each component was manually inspected for flight readiness before launch testing.

At a live launch event, all four rockets achieved stable launches, and two were recovered in re-flyable condition—a notable success given the complexity. Flight data analysis revealed that RocketSmith's simulations predicted apogee with 84% accuracy when compared to measured values. This demonstrates the system's robust capability to go from conceptual design to physical testing autonomously. The work, published on arXiv (cs.RO), showcases how agentic AI can streamline aerospace prototyping, potentially reducing design cycles and enabling rapid experimentation in rocketry and other engineering domains.