PaperlayerView exact revision on arXiv

Source-linked AI summary

RocketSmith: Agentic Additive Manufacturing of High-Powered Rockets

Peter Pak, Jesse Barkley, Rumi Loghmani, Derek Baich, Ananya Pamal, Amir Barati Farimani

arXiv:2606.00097v3cs.ROcs.MA

TL;DR

High-powered rocket development is iterative and time-consuming, motivating automation of design and manufacturing workflows. RocketSmith orchestrates software tools and additive manufacturing to develop rockets, with all four tested rockets launching stably and apogee accuracy reaching 80% and 84% for two rockets.

  • Problem

    High-powered rocket development involves iterative, time-consuming design and manufacturing processes.

  • Method

    RocketSmith orchestrates software tools and additive manufacturing to automate high-powered rocket design and development.

  • Results

    All four rockets launched stably, two were recovered reflyable, and apogee accuracy was 80% for High Power 1 and 84% for High Power 4.

  • Takeaways & Limitations

    Flight tests demonstrate RocketSmith’s automated development of additively manufactured high-powered rockets.

  • Takeaways & Limitations

    The development workflow retains human-in-the-loop considerations.

Abstract

from arXiv · show

RocketSmith is an agentic system which intelligently automates the DFAM process for the development of high powered rockets suitable for launch. The system utilizes a large language model to orchestrate the execution of software tools to validate design characteristics such as flight stability and generate the parametric design components for the rocket assembly. A collection of subagents and skills enable optimization workflows of flight parameters via iteration in both zero-shot and human-in-the-loop workflows. With this system, four distinct high power rockets with various motor and assembly configurations were developed utilizing the unique design capabilities of additive manufacturing. These assembly components were fabricated using various FDM printers, manually evaluated for flight readiness, and flight tested at a launch event. From these tests, all rockets achieved a stable launch and two of the four rockets were successfully recovered in reflyable condition. The altimeter data validated that the rockets achieved an altitude 80% of the expected apogee predicted by the agentic system, establishing consistency between simulation and experimentation.

1 Introduction

RocketSmith is introduced as an agentic system that streamlines the iterative design, simulation, CAD, and manufacturing workflow for high-powered rockets. It orchestrates subagents, skills, and MCP-based tool calling under user constraints, then supports additive fabrication and flight-test evaluation.

  • Motivation: Additive manufacturing is presented as a distinctive solution for part consolidation and flight optimization in the multifaceted design and manufacturing of high-powered rockets.The introduction frames aerospace as a development-constrained and cost-elastic industry where additive manufacturing is attractive.
  • Motivation: Rocket development is iterative because manufacturing constraints, part availability, and unforeseen events require recalculating stability and revising assembly decisions.The introduction identifies friction between software tools for flight simulation, design, and manufacturing as a problem in this loop.
  • System: RocketSmith combines subagents, skills, and Model Context Protocol tool calling and is released as a plugin for agent harnesses such as Claude Code.Its large language model supplies domain-specific reasoning and tool-interaction capabilities for iterative development.
  • System: Under user-provided constraints and specifications, RocketSmith orchestrates flight simulations, CAD-file design, and manufacturing-file generation for high-powered rockets.The system can design an OpenRocket-based component tree, generate parametric airframe models with CADSmith, and improve center-of-gravity estimates using PrusaSlicer weights.
  • Evaluation: RocketSmith schematics are additively manufactured via Fused Deposition Modeling and assembled for evaluation through flight tests at a launch event.Quantitative altimeter data from a subset of flights and qualitative metrics are used to evaluate the agentic system’s general performance.

2 Related Works

Related work demonstrates agentic systems for improving additive manufacturing and evaluating rocket design, while exposing limitations in LLM-driven iterative optimization and text-to-CAD precision. RocketSmith addresses these limitations by delegating numerically precise engineering tasks to tools and structuring verifiable workflows.

  • Rocket-design evaluation: RocketBench connects LLMs to RocketPy trajectory simulation and scores structured JSON rocket designs using altitude accuracy, structural integrity, drift, cost, and landing safety.Frontier LLMs show strong baseline engineering intuition but plateau below human expert performance during iterative refinement.
  • RocketSmith motivation: RocketBench identifies a failure mode in which standard LLMs struggle to translate simulation feedback into meaningful design updates.RocketSmith therefore delegates stability calculation, trajectory simulation, slicing, and mass estimation to numerically precise tools.
  • Text-to-CAD systems: Early text-to-CAD methods demonstrated geometric reasoning but accumulated dimensional errors because modeling sequences lacked external correction signals.Agent-Aided Design systems instead use feedback loops that generate, compile, and visually inspect geometry.

3 Background

Background establishes the regulatory scope of high-power rocketry, the stability and manufacturing constraints governing successful flight, and the agentic-system concepts used to automate complex workflows. It also describes how LLM reasoning, tool calls, and runtime harnesses support task decomposition and validation.

  • High-Power Rocketry: Class 2 high power rocketry permits total impulse up to 40,960 N · s, while Class 3 covers rockets beyond 40,960 N · s and up to 889,600 N · s.Access is governed by certifications from TRA and NAR, with flight waivers administered by the FAA.
  • Rocket Design Constraints: A stability value between 1.00 cal and 1.25 cal is desired because low stability can cause tumbling, whereas excessive stability promotes weathercocking.Stability is determined from center of pressure, center of gravity, and body tube diameter.
  • Design and Manufacturing: Design, manufacturing, and assembly must account for launch and recovery events, weight, accessibility, manufacturing, materials, tolerances, and measured flight parameters.Actual center-of-gravity measurements from assembled rockets can be used to adjust flight simulations.
  • Agentic Systems: An agentic system uses LLM reasoning, context, tool calls, and optimization to decompose natural-language goals into subtasks and validate results.Its principal components are LLM-enabled orchestration, precise tool calls, and an agent harness providing runtime support.
  • Tools and Agent Harnesses: Tool calling lets LLMs interact with external environments through predetermined functions, while the agent harness manages inference, tool execution, conversation history, context windows, and termination.Tool-call failures can include schema mismatches and semantic errors, with MCP presented as a standard protocol for avoiding them.

4 Methodology · 4.1 Software Tools · 4.2 Agentic System

RocketSmith combines OpenRocket, build123d, and PrusaSlicer in a downstream design-to-manufacturing pipeline, orchestrated by an agentic system supporting zero-shot and human-in-the-loop workflows. Its subagents and skills coordinate simulation, parametric CAD generation, slicing, visualization, and manufacturing preparation.

  • 4.1 Software Tools: RocketSmith’s core pipeline uses OpenRocket, build123d, and PrusaSlicer to automate component design and manufacturing.OpenRocket generates simulations, build123d generates parametric STEP files, and PrusaSlicer estimates component weights and produces tool paths.
  • 4.1.1 OpenRocket: OpenRocket is RocketSmith’s primary source of truth, defining dimensional, weight, and assembly configurations before downstream tools and supporting flight simulations.Its simulations use component properties and variables such as wind conditions, recovery deployment, and launch rail length to estimate altitude, velocity, acceleration, stability, and apogee.
  • 4.1.2 build123d: After OpenRocket verifies stability and flight readiness, build123d generates a Python script and STEP file for every printable component.The parametric workflow permits targeted edits, programmatic extraction of bounding box, volume, and mass, and downstream use in PrusaSlicer or other tooling.
  • 4.1.3 PrusaSlicer: PrusaSlicer estimates STEP-component weight using expected material, print, and printer configurations, then uploads generated tool paths for fabrication.Fabricated parts are subsequently weighed to obtain actual values.
  • 4.2 Agentic System: The agentic system augments rocket development through zero-shot and human-in-the-loop operation, with a GUI for monitoring CAD models, simulations, and component trees.Subagents compartmentalize specialized context, while skills provide concrete workflows for recurring tool-use behaviors.
  • 4.2.1 Graphical User Interface (GUI): The GUI provides a readonly visualization of RocketSmith’s current state, including component trees, flight simulations, CADSmith scripts and models, and assembly configurations.The primary execution interface is the CLI-based agent harness of Claude Code.
  • 4.2.2 Subagents: The system comprises 6 subagents that divide responsibilities across orchestration, CAD, GUI, manufacturing, OpenRocket, and PrusaSlicer.The manufacturing subagent manages OpenRocket-to-CADSmith handoffs and evaluates DFAM and DFM adjustments.
  • 4.2.3 Skills: RocketSmith instantiates 7 skills spanning motor selection, stability analysis, structure generation and modification, additive manufacturing, mass calibration, and print preparation.Motor selection starts workflows, stability analysis verifies the OpenRocket design, and print preparation sends STEP files to PrusaSlicer for weight estimation and tool-path generation.

4.3 Manufacturing · 4.4 Assembly

Four high-powered rockets were manufactured by FDM printing in PETG or ABS with varied printer, material, and segmentation choices, then assembled into distinct motor and recovery configurations. The assembly designs ranged from two-part to three-part airframes and incorporated rocket-specific attachment, ejection, instrumentation, and recovery features.

  • 4.4 Assembly: The four rockets were designed and assembled by Pak, Loghmani, and Barkley across level 1 and level 2 impulse classifications.Pak designed High Power 1 and High Power 4; Loghmani designed High Power 2; and Barkley designed High Power 3.
  • 4.4.1 High Power 1: High Power 1 was a single motor deploy level 1 rocket with a 38 mm AeroTech H100W motor, 36” parachute, ABS lower airframe, and PETG nose cone.Slots housed a StratoLogger CF altimeter and RunCam 5 camera for flight data recording.
  • 4.4.4 High Power 4: High Power 4 was the largest rocket, a level 2 impulse design using an AeroTech J425R motor, 800 mm lower airframe, 1150 mm overall height, and 48” parachute.It included slots for a StratoLogger CF altimeter and RunCam 5 camera, with two steel eyebolts and kevlar shock cord connecting the airframe and nose cone.

5 Results

Flight tests were conducted under nearly ideal conditions with preflight checks and range-safety oversight. All four rockets launched stably; two were recovered reflyable, while instrumented flights reached 80% and 84% of expected apogees.

  • Flight testing: All four rockets underwent supervised preflight checks, and suggested adjustments were applied before flight testing.Testing occurred under clear skies and moderate winds around 10 mph.
  • Overall results: Two of four rockets, High Power 3 and High Power 4, were recovered in reflyable condition.High Power 1 experienced structural failure, while High Power 2 experienced recovery deployment failure.
  • High Power 1: 80%: High Power 1 reached 276 m (905 ft) against an expected apogee of 338 m (1108 ft).Its electronics recorded flight data despite a catastrophic lower-airframe structural failure that left the rocket non-reflyable.
  • High Power 2: High Power 2 launched vertically and stably, but recovery-bay separation failed and the rocket impacted the ground at high velocity.The rocket embedded approximately 4 meters into the soil, although its fins remained intact.
  • High Power 4: 84%: High Power 4’s altimeter measured an apogee of 479 m (1571 ft) against an expected 570 m (1870 ft).Recovery-bay separation and 48” parachute deployment succeeded, and the rocket was recovered in reflyable condition.

6 Discussion

RocketSmith automated much of high-powered rocket design and manufacturing, reducing development time and friction, but human guidance and external design or manufacturing steps remained necessary. Flight tests showed stable launches and close agreement between manufactured rockets and agentic-system designs, while recovery performance depended on implementation details.

  • System limitations: Manual intervention remained necessary: some CAD components, complex covers, and the High Power 4 motor retainer were designed outside RocketSmith.Actual slicing and printing were also performed externally with printer-specific profiles, partly to reduce print-failure risk before launch.
  • Manufacturing limitations: ABS cracking occurred across airframe parts, especially High Power 1 and High Power 4, and was repaired after fabrication with epoxy.Proposed remedies included improved chamber insulation, less temperature-sensitive material such as PETG, and future PrusaSlicer visual validation for a truly end-to-end pipeline.
  • Development workflow: RocketSmith significantly reduced the time and friction of developing and testing high-powered rockets, although human-in-the-loop use remained beneficial alongside feasible zero-shot operation.The most time-consuming aspect was repeated design and simulation iteration for stability and recovery deployment.
  • Flight testing: High Power 4 reached approximately 84% of expected apogee, followed the expected trajectory and descent rate, and was successfully recovered intact in reflyable condition.It performed best among the tested rockets but involved the most human participation during development.

7 Conclusion

RocketSmith automated the development of additively manufactured high powered rockets through agentic subagents, skills, CAD generation, and iterative flight optimization. Flight testing showed stable launches, successful recovery of two rockets, and measured apogee accuracy for High Power 1 and High Power 4.

  • 7 Conclusion: RocketSmith automated high powered rocket development by outlining blueprints, designing for additive manufacturing, generating CAD models, and optimizing flight-critical values iteratively.Subagents and skills supported the agentic workflow.
  • 7 Conclusion: Two of the four tested high power rockets were successfully recovered in reflyable condition after flight testing with FDM-manufactured outputs.The rockets were flight tested at a launch event.
  • 7 Conclusion: All rockets achieved a stable launch, while onboard instrumentation showed apogee accuracy of 80% for High Power 1 and 84% for High Power 4.These values were reported for the respective tested rockets.
  • 7 Conclusion: The results demonstrate that an agentic system can design complex assemblies suitable for additive manufacturing in high powered rocketry.The conclusion identifies high powered rocketry as a domain-specific validation application.

Appendix A Motor Impulse Classification

Motor classification is based on total impulse, defined as thrust integrated over burn time, with rocket motor codes specifying impulse, average thrust, and delay information. FAA classifications distinguish Model, High Power, and Advanced High Power Rockets, while Class 2 rockets are limited to 40,960 Newton-seconds of combined impulse.

  • Impulse Basis: Motor classification is determined by the total impulse delivered by the propulsion system.Total impulse is established by integrating thrust over the motor’s burn duration.
  • Impulse Basis: Thrust force F is defined as propellant mass flow rate multiplied by exhaust velocity.Mass flow rate is measured in kg/s and exhaust velocity in m/s.
  • Rocket Classes: FAA rocket classes comprise Class 1 Model Rocket, Class 2 High Power Rockets, and Class 3 Advanced High Power Rockets.These classifications apply specifically to amateur rocket activities.
  • Motor Codes: Rocket motor codes specify total impulse by letter, average thrust by the intermediate number, and recovery-ejection delay by the final number.For example, H100W-14 indicates 160.01–320 Newton-seconds total impulse, 100 newtons average thrust, and a 14-second maximum delay.
  • Class 2 Requirements: 40,960 Newton-seconds is the combined total impulse limit for most Class 2 High Power Rocket activities, spanning H to O motors.Certification is required to purchase and launch high powered rockets.

Appendix B Voron-2-Tall

Voron-2-Tall is a modified Voron 2.458 FDM printer with an extended z-axis build height suited to printing tall rocket airframe parts. Its original 930 mm maximum build height was reduced to 850 mm by frame braces.

  • Printer configuration: Voron-2-Tall is a modified Voron 2.458 FDM printer with a maximum build height of 930 mm for tall rocket airframe parts.The printer was built from a Voron 2.458 kit and extended along the z axis.
  • Printer configuration: 850 mm is the build height limit after frame braces reduced the original 930 mm maximum.Custom 2020 aluminum extrusions, MGN9H linear rails, GT2 belts, and enclosure panels extended the z axis over a 1000 mm range before bracing.
  • Printer configuration: The printer also incorporated a DragonBurner toolhead and CANBUS to reduce form factor and wiring.These modifications were made in addition to the extended z-axis structure.
  • Print validation: A successful Benchy print demonstrated operation with PLA, while a lower airframe component for Big Heavy 2 was printed using ABS.Figure 18 shows the assembled printer, its first successful Benchy print, and the finished lower airframe component.

Appendix C Onboard Camera Data

Onboard RunCam 5 cameras captured launch-site views from High Power 1 just before an apogee of 338 m (1108 ft) and from High Power 4 around a measured apogee of 570 m (1870 ft).

  • Onboard Camera Data: 338 m (1108 ft): High Power 1’s onboard RunCam 5 captured the launch site just before reaching apogee.The image was taken from the rocket before it reached the stated apogee.
  • Onboard Camera Data: 570 m (1870 ft): High Power 4’s RunCam 5 captured a photo around its measured apogee.The photograph was taken using the onboard camera near the measured apogee.

Appendix D Launch Site Weather Conditions

Table 4 records the hourly weather conditions during the Sunday, May 3, 2026 launch window at Fayette County, Pennsylvania.

  • Launch Site Weather Conditions: Table 4 presents hourly weather data for the Sunday, May 3, 2026 launch window at Fayette County, PA.The passage identifies the launch date, location, and hourly temporal coverage but provides no individual weather values.
Loading 2606.00097v3…