Mechanism constraint structures improve learning from limited data
Efficient Learning of Mechanism Designs and Compositionality
Artificial IntelligenceComputational Engineering, Finance, and Science
Summary
Predicting how mechanical parts move together is hard when data is limited. The authors study if breaking mechanisms into reusable pieces helps learning. They create a method called Assur-KB that captures these pieces and their connections using geometric and constraint information. Their experiments show this approach learns better and generalizes well with fewer samples than simpler methods. This work helps understand which physical connections matter most for reusing parts in learning mechanical behavior.
What this means in practice
- •For mechanical design engineers: Improve predictive models of mechanical systems from small datasets by using modular constraint representations.
- •For robotics developers: Generate valid new mechanism designs more reliably by composing learned modules with consistent interface conditions.
Authors
João Pedro Gandarela, Thiago Rios, Stefan Menzel, André Freitas
Abstract
Learning mechanism behaviour requires a representation of how connected parts jointly constrain motion. We study whether exposing reusable constraint subassemblies improves prediction from limited data. Assur-KB represents planar mechanisms through irreducible modules, boundary interfaces, and solve dependencies while retaining source geometry. Our theoretical results characterize transfer through module interfaces and establish physical limits arising from missing interface information and accumulated sensitivity. To test the representation claim directly, the experimental design compares matched mechanism-curve pairs across training budgets within the same multilayer-perceptron family. This controlled comparison establishes the utility of the complete representation in the evaluated learning setting. Assur-derived features provide a geometric-mean $1.8\times$ advantage in sample efficiency over flattened adjacency and position inputs across budgets of 50-10,000 mechanisms. Separate experiments establish invariance under tested joint relabelings, higher valid-generation rates under dyadic composition, and a coarser structural index over 10.02 million mechanisms. In summary, the theory and experiments identify how physical constraint structure supports limited-data prediction and which interface conditions govern compositional reuse.