Explore new structures
Our systems generate geometry programs that explore new topologies, channel layouts, lattices, and manifolds.
Vinland Labs / Physical invention
AI systems that invent new physical technologies.
Vinland Labs develops AI systems that discover novel physical technologies through simulation and open-ended search. We focus on new hardware designs and licensable inventions, starting with spacecraft cooling hardware.
Loading generated geometry
Vortex-core heat exchanger
Voxel-native / 3 mesh bodies / drag to inspect
Current research
01
Spacecraft cooling
02
Physics simulation
03
Open-ended search
The problem
The design space is too large for humans to search.
Novel physical technologies require exploring beyond familiar designs. Conventional optimization starts with a predefined family of shapes. We develop AI systems to search across new structures, using physics simulation to evaluate what they discover.
Conventional search
Predefined design families
Unsearched structures
Unexplored hardware designs
Explored
Leif search
Our research connects open-ended search with physics simulation to discover new hardware.
Evaluate through simulation
Physics simulation evaluates candidate designs against performance targets and physical constraints, guiding the next round of search.
Build on discoveries
Useful design patterns become building blocks for further exploration, expanding the structures our systems can generate.
Our approach
An inventor that learns through search.
Leif is the invention system we are developing. Our research combines program synthesis, library learning, and deep learning to generate new hardware designs, evaluate them in simulation, and use discoveries to guide further search.
01
Program synthesis
Generate hardware designs
Geometry programs describe how a design is constructed, allowing the search to explore different structures and arrangements of physical components.
02
Library learning
Learn reusable structures
Library learning captures useful geometry patterns as reusable concepts that can be combined into new designs.
03
Deep learning
Choose where to search
Learned models guide the search toward promising candidates for further simulation and eventual physical validation.
The research goal
Discoveries that lead to new inventions.
We are building a system that carries discoveries forward, recombining useful concepts to explore new physical technologies. Our intended output is novel hardware and licensable inventions.
Our first focus
Inventing spacecraft cooling hardware.
We are starting with spacecraft cooling hardware, exploring new designs for heat exchangers, cold plates, manifolds, and internal flow channels.
Our search evaluates how new geometries manage heat within mass, volume, flow, and manufacturing constraints. We aim to turn promising discoveries into validated hardware designs and inventions that industry partners can license.
Maximum temperature
Heat transfer
Pressure drop
Mass
Packaging
Manufacturability
Research & licensing
Bring new inventions into the world.
We welcome conversations with partners interested in spacecraft cooling, physical validation, and licensing new hardware inventions.
Work with Vinland Labs