Recently Published by Autodesk Researchers

Autodesk Research teams regularly contribute to peer-reviewed scientific journals and present at conferences around the world. Check out some recent publications from Autodesk Researchers.

DiffVL: Scaling up Soft Body Manipulation using Vision-Language Driven Differentiable Physics

Combining gradient-based trajectory optimization with differentiable physics simulation is an efficient technique for solving soft-body manipulation problems. Using a well-crafted optimization objective, the solver can quickly converge onto a valid trajectory. However, writing the appropriate objective functions requires expert knowledge, making it difficult to collect a large set of naturalistic problems from non-expert users. In this paper, the team introduces DiffVL, a method that enables non-expert users to communicate soft-body manipulation tasks a combination of vision and natural language, given in multiple stages – that can be readily leveraged by a differential physics solver.

Generative Design through Quality Diversity Data Synthesis and Language Models

Two fundamental challenges face generative models in engineering applications: the acquisition of high-performing, diverse datasets, and the adherence to precise constraints in generated designs. In this research, the team proposes a novel approach combining optimization, constraint satisfaction, and language models to tackle these challenges in architectural design. Their method uses Quality-Diversity (QD) to generate a diverse, high-performing dataset that they then fine-tune a language model to generate high-level designs. These designs are then refined into detailed, constraint-compliant layouts using the Wave Function Collapse algorithm.

HG-CAD Hierarchical Graph Learning for Material Prediction and Recommendation in Computer-Aided Design

To support intelligent computer-aided design (CAD), the team introduces a machine learning architecture, namely HG-CAD, that recommends assembly body material through joint learning of body and assembly-level features using a hierarchical graph representation. Specifically, the team formulates the material prediction and recommendation process as a node-level classification task over a novel hierarchical graph representation of CAD models, with a low-level graph capturing the body geometry, a high-level graph representing the assembly topology, and a batch-level mask randomization enabling contextual awareness. This enables the network to aggregate geometric and topological features from both the body and assembly levels, leading to competitive performance.