Search papers, labs, and topics across Lattice.
This paper introduces AADvark, an agentic system for designing dynamic CAD models with moving parts, a capability lacking in previous Agent-Aided Design systems. AADvark addresses the spatial reasoning limitations of LLMs by integrating external constraint solvers and a specialized visual feedback mechanism. Experiments demonstrate AADvark's ability to generate 3D assemblies with movable parts by leveraging FreeCAD and an assembly solver to create a strong verification signal.
Agent-Aided Design can now create functional assemblies with moving parts, like pistons and pendulums, by augmenting LLMs with constraint solvers.
In the past year, researchers have started to create agentic systems that can design real-world CAD-style objects in a training-free setting, a new variety of system that we call Agent-Aided Design. Generally speaking, these systems place an agent in a feedback loop in which it can write code, compile that code to an assembly of CAD model(s), visualize the model, and then iteratively refine its code based on visual and other feedback. Despite rapid progress, a key problem remains: none of these systems can build complex 3D assemblies with moving parts. For example, no existing system can build a piston, a pendulum, or even a pair of scissors. In order for Agent-Aided Design to make a real impact in industrial manufacturing, we need a system that is capable of generating such 3D assemblies. In this paper we present a prototype of AADvark, an agentic system designed for this task. Unlike previous state-of-the-art systems, AADvark captures the dynamic part interactions with one or more degrees-of-freedom. This design decision allows AADvark to reason directly about assemblies with moving parts and can thereby achieve cross-cutting goals, including but not limited to mechanical movements. Unfortunately, current LLMs are imperfect spatial reasoners, a problem that AADvark addresses by incorporating external constraint solver tools with a specialized visual feedback mechanism. We demonstrate that, by modifying the agent's tools (FreeCAD and the assembly solver), we are able to create a strong verification signal which enables our system to build 3D assemblies with movable parts.
