Winter 2027 Research Engineer Co-op

Maritime Digital Production is the software and digital systems function within Anduril's Heavy Metal division. We build and deploy the technology stack that powers Anduril's shipbuilding factories: the data infrastructure that makes every machine and sensor visible in real time, the manufacturing execution system that workers and planners use every shift, the scheduling engine that replans production in minutes instead of days, and the AI systems that remove manual toil from the shop floor and business operations. That includes Serapis, our program building components for the Virginia-class submarine.

We operate at the boundary between Operational Technology and Information Technology, incubating solutions close to the production line, validating them with real operators, hardening them for reliability and security, and scaling them across sites. The environment is fast, physical, and consequential. When our systems go down, production stops. The output of our work is not a dashboard: it is a ship.

We are scaling from a founding team to more than 70 engineers across multiple U.S. sites. As an intern you join early, on a problem that matters to production, with a defined project and a named engineer you work with every day.

We have built a computational geometry engine that takes a 3D CAD assembly and works out the orders in which it can physically come apart, and therefore the orders in which it can be built. It finds where parts touch, builds directional blocking graphs from that contact geometry, searches for collision-free removal directions, and produces the precedence constraints our production scheduling system plans against.

It works on small and mid-sized assemblies. It needs to work on the assemblies we actually build, which have hundreds to thousands of parts. Your job is to close that gap.

This is algorithms engineering with a research flavor. You will profile the engine on real assemblies, find where the time and memory go, and attack it: search ordering, pruning, memorization, incremental updates, subassembly decomposition, and parallelism. The goal is a measurably faster engine rather than a paper, and every result is held to one rule: a faster answer that is a different answer is a bug.

WHAT YOU’LL DO

• Profiling a C++ computational geometry pipeline on real assemblies and reporting where time and memory go
• Designing and implementing algorithmic improvements to a disassembly sequence search: candidate ordering, pruning, memoization, and incremental graph updates
• Decomposing large assemblies into subassemblies that can be solved independently
• Parallelizing expensive geometric tests across cores without losing determinism
• Building a benchmark corpus and harness that run in continuous integration and catch regressions
• Gating every optimization on output equivalence against a reference set
• Writing up results and presenting them to the engineering team

REQUIRED QUALIFICATIONS

• Must be pursuing master’s degree or PhD in Computer Science, Robotics, Mechanical Engineering, Applied Mathematics, or a related field.

• Must be returning to school at the end of the internship to continue academic studies for at least one quarter or semester.
• Must be a U.S. Person due to required access to U.S. export-controlled information or facilities; U.S. clearance eligibility may be required depending on program.
• Ability to complete an in-person co-op in Santa Ana, CA, in winter 2027.

PREFERRED QUALIFICATIONS

• Strong modern C++, C++17 or later. You have written and debugged performance-sensitive C++ in a codebase you did not write yourself, not only coursework.
• Solid algorithms and data structures, including graph algorithms such as strongly connected components and topological ordering, combinatorial search and pruning, and complexity analysis.
• Computational geometry fundamentals: triangle meshes, bounding volume hierarchies and other spatial indexes, intersection and collision tests, and reasoning about directions on the sphere.
• Experience profiling and optimizing code. You have used a profiler and made something measurably faster.
• Research or coursework in assembly sequence planning, disassembly planning, or motion planning. If you know what a non-directional blocking graph is, or have read Wilson and Latombe or the STAAT work by Romney and colleagues, we want to talk to you.
• Publications at ICRA, IROS, RSS, SoCG, or Solid and Physical Modeling, or in the Computer-Aided Design journal.
• Experience with Eigen, OpenCASCADE, OpenVDB, CGAL, or a collision or bounding volume hierarchy library.
• Parallel programming in C++ with Intel TBB, OpenMP, or standard threads, with attention to determinism.
• Familiarity with CAD data: boundary representations versus tessellated meshes, and formats such as STEP and JT.
• Experience with CMake, GoogleTest, and reproducible builds. Nix is a plus.
• Python for tooling and analysis, e.g. NumPy and trimesh.
• Familiarity with infinitesimal motion analysis or rotational motions in assembly planning.