What if a CAD system could automatically generate tens, hundreds, or even thousands of design options that all meet your specified high-level goals? It’s no longer what if: it’s Project Dreamcatcher, and it’s the next generation of computational design.
Dreamcatcher is a generative design system that enables designers to input specific design objectives, including functional requirements, material type, manufacturability, performance criteria, and cost restrictions. The infinite computing power of the cloud then takes over.
Recent advancements in artificial intelligence and the simulation of complex phenomena have enabled software to play an active, participatory role in the invention of form. Project Dreamcatcher is an experimental design platform with focused research probes into generative design systems.
The Dreamcatcher system allows designers to input specific design objectives, including functional requirements, material type, manufacturing method, performance criteria, and cost restrictions. Loaded with design requirements, the system then searches a procedurally synthesized design space to evaluate a vast number of generated designs for satisfying the design requirements. The resulting design alternatives are then presented back to the user, along with the performance data of each solution, in the context of the entire design solution space.
Upon evaluating the progressively generated solutions, the designer is then able to return to the problem definition at any time to adjust goals and constraints to generate new results that fit the refined definition of success. Once the design space has been explored to satisfaction, the designer is able to output the design to fabrication tools or export the resulting geometry for use in other software tools.
Dreamcatcher includes two key software components for capturing and logically structuring users' design inspiration into machine-readable goals and constraints. Dreamcatcher's problem structure is a language for designers to describe design problems. Through pattern-based description, solutions become modular and accretive, thereby expanding the quality and number of alternatives that are searched in a Dreamcatcher design session. The problem graph is a software tool for visualizing and interacting with the problem structure. The Dreamcatcher design ontology, created through machine learning techniques,is a classified index of pre-existing objects that perform functions, or satisfy constraints, similar to those the user has defined in their problem definition. This ontology enables Dreamcatcher to reason by analogy.
Mimicking the variety of reference material in a typical design brief, in Dreamcatcher the designer explicitly and implicitly documents goals and constraints through a number of input modalities including natural language, image inference and CAD geometry. An individual or team may manipulate the problem definition through these multiple modes of input and verify or modify the inferred changes to the problem definition document. Change propagation may occur from the top-down or bottom-up and alterations are visualized through the problem graph.
The Dreamcatcher team is developing and testing several, purpose-built geometry representations that alternatively optimize for shape complexity, analysis speed, simulation accuracy, etc. of procedurally generated geometry. For example, the level set representation for 3D geometry allows for infinite shape and material complexity. Through creating simulation tools that integrate with level sets rather than mesh geometries, geometry synthesis from simulation results is more expedient and reliable. The Dreamcatcher system enables designers to truly leverage an emerging class of manufacturing tools that release designers from hundreds of years of predicating design decisions on tool based constraints.
Through a purpose-built, scalable and parallelized cloud computing framework code named Saturn, Dreamcatcher is able to generate and evaluate solution sets with complexity well beyond that of Generative Design Systems of the past. Saturn provides the high-performance computing infrastructure necessary to run the computationally intense optimization and analysis engines, including multi-physics simulations.
After a number of solutions have been computationally generated from a problem definition, the Dreamcatcher design explorer presents to the user a set of possible solutions and their associated solution strategies. This user interface provides a sense of the shape of the valid design space and variable interactions. It also assists users in building a mental model of which alternatives are high performing relative to all others in the set. Once the solution has been adequately explored, the designer can modify the problem definition to iteratively generate more relevant solutions.
Traditional optimization workflows like that of the NASA ST-5 antenna are 'closed loop' where a design space is defined and then searched by a genetic algorithm or similar optimization function. Arguments for the incorporation of AI into design often default to concerns around replacing the human designer. While many elements that are commonly modeled from scratch such as brackets, adapters and stiffeners may be created more effectively by a system such as Dreamcatcher, complex elements and aspects that are difficult to quantify will require new types of interaction to leverage human intuition and computational rigor in partnership. Dreamcatcher is pioneering new methods for interactive synthesis and optimization with industry leaders from the automotive, aerospace and manufacturing fields.
The Dreamcatcher team consists of the Computational Science and Design Research groups of Autodesk Research in the Office of the CTO with collaborators throughout the larger Autodesk Corporation, industrial partners and academic partners. The global team in San Francisco, Toronto, and London UK is made up of specialists from varied domains as mathematical optimization, geometry, machine learning, mechanical engineering, material science, structural mechanics, user experience research, software design and development.
To learn more about Autodesk's initiatives in generative design, please visit our generative design information page.
Vincent Goulet, Wei Li, Hyunmin Cheong, Francesco Iorio, Claude-Guy Quimper (2016)Four-Bar Linkage Synthesis Using Non-Convex Optimization
Hyunmin Cheong, Wei Li, Francesco Iorio (2016)Automated Extraction of System Structure Knowledge from Text
Xiaoxiao Guo, Wei Li, Francesco Iorio (2016)Convolutional Neural Networks for Steady Flow Approximation
Hyunmin Cheong, Wei Li, Adrian Cheung, Andy Nogueira, Francesco Iorio (2015)Automatic Extraction of Function Knowledge from Text
Hyunmin Cheong, Wei Li, Francesco Iorio (2015)A Novel Application of Gamification for Collecting High-Level Design Information
Michael Bergin, Mohammad Rahmani Asl, Adam Menter, Wei Yan (2014)BIM-based Parametric Building Energy Performance Multi Objective Optimization
Hyunmin Cheong, Wei Li, Li Shu, Erin Bradner, Francesco Iorio (2014)Use of Controlled Natural Language to Input Problem Definition for Computer-Aided Design
Erin Bradner, Francesco Iorio, Mark Davis (2014)Parameters Tell the Design Story: Ideation and Abstraction in Design Optimization
Hyunmin Cheong, Wei Li, Li Shu, Alex Tessier, Erin Bradner, Francesco Iorio (2014)Natural Language Problem Definition for Computer-Aided Mechanical Design