US-12619795-B2 - Artificial intelligence-based techniques for design generation in virtual environments

Abstract

A technique for generating designs includes: causing one or more candidate designs to be displayed within a virtual-reality (VR) environment; receiving a user input associated with a first candidate design included in the one or more candidate designs via the VR environment; generating a modified design based at least on the user input and the first candidate design; and generating a plurality of output designs via a generative design process based on the modified design.

Inventors

• Fraser ANDERSON
• Josh Davis
• George Fitzmaurice
• Tovi Grossman
• Merten Stroetzel

Assignees

• AUTODESK, INC.

Dates

Publication Date

20260505

Application Date

20210824

Claims (20)

1. 1 . A computer-implemented method for generating designs, the method comprising: displaying one or more candidate designs within a virtual-reality (VR) environment; receiving a user input indicating a modification of a mesh of a first candidate design included in the one or more candidate designs via the VR environment; generating a modified design based on the modification of the mesh of the first candidate design; generating, by a generative design engine, a plurality of output designs based on the modified design, wherein the generative design engine freezes weights associated with the generative design engine, identifies a vector in a latent space that represents a voxelated version of the user-modified design, and modifies the voxelated version of the modified design to generate each of the output designs; and displaying at least one output design included in the plurality of output designs within the VR environment.
2. 2 . The computer-implemented method of claim 1 , wherein the user input indicating the modification of the mesh associated with the first candidate design further indicates a combination of the first candidate design and a second candidate design included in the one or more candidate designs, a region of the first candidate design that is changeable, or a region of the first candidate design that is not changeable.
3. 3 . The computer-implemented method of claim 1 , wherein the user input further indicates a combination of the first candidate design and a second candidate design, and wherein generating the modified design comprises performing a linear interpolation between the first candidate design and the second candidate design.
4. 4 . The computer-implemented method of claim 3 , wherein the linear interpolation is based on a first z-vector sampled from a probabilistic latent space associated with the first candidate design and a second z-vector sampled from a probabilistic latent space associated with the second candidate design.
5. 5 . The computer-implemented method of claim 1 , wherein the user input further indicates a combination of the first candidate design and a second candidate design, and wherein the first candidate design is associated with a first object class, and the second candidate design is associated with a second object class.
6. 6 . The computer-implemented method of claim 5 , wherein the first object class is associated with the plurality of output designs generated by the generative design engine, and the second object class is associated with objects that are not generated by the generative design engine.
7. 7 . The computer-implemented method of claim 5 , wherein the first object class is associated with a first category of the plurality of output designs generated by the generative design engine, and the second object class is associated with a second category of the plurality of output designs generated by the generative design engine.
8. 8 . The computer-implemented method of claim 1 , further comprising, prior to displaying the one or more candidate designs, converting the one or more candidate designs from a voxel-based format to a mesh-based format.
9. 9 . The computer-implemented method of claim 1 , wherein the user input comprises a user gesture within the VR environment.
10. 10 . The computer-implemented method of claim 1 , further comprising, prior to displaying the one or more candidate designs, generating a plurality of object designs via the generative design engine based on an initial design.
11. 11 . The computer-implemented method of claim 1 , wherein the user input comprises receiving, via one or more VR controllers, a user gesture for sculpting a surface mesh within the VR environment.
12. 12 . The computer-implemented method of claim 1 , wherein the user input indicating the modification of the mesh is performed relative to a region of the first candidate design that is not changeable by the generative design engine that generates the modified design.
13. 13 . A non-transitory computer readable medium storing instructions that, when executed by a processor, cause the processor to perform the steps of: displaying one or more candidate designs within a virtual-reality (VR) environment; receiving a user input indicating a modification of a mesh of a first candidate design included in the one or more candidate designs via the VR environment; generating a modified design based on the modification of the mesh of the first candidate design; generating, by a generative design engine, a plurality of output designs based on the modified design, wherein the generative design engine freezes weights associated with the generative design engine, identifies a vector in a latent space that represents a voxelated version of the user-modified design, and modifies the voxelated version of the modified design to generate each of the output designs; and displaying at least one output design included in the plurality of output designs within the VR environment.
14. 14 . The non-transitory computer readable medium of claim 13 , wherein displaying the one or more candidate designs comprises positioning the one or more candidate designs within a first virtual display region of the VR environment.
15. 15 . The non-transitory computer readable medium of claim 14 , wherein the first virtual display region includes a different virtual display volume for each of the one or more candidate designs.
16. 16 . The non-transitory computer readable medium of claim 14 , wherein the first virtual display region includes at least one virtual display volume for the modified design.
17. 17 . The non-transitory computer readable medium of claim 14 , wherein the steps further include displaying the modified design within a virtual display volume included in the first virtual display region.
18. 18 . The non-transitory computer readable medium of claim 14 , wherein the at least one output design included in the plurality of output designs is displayed within a second virtual display region of the VR environment.
19. 19 . The non-transitory computer readable medium of claim 18 , wherein the second virtual display region is adjacent to the first virtual display region.
20. 20 . The non-transitory computer readable medium of claim 18 , wherein the first virtual display region and the second virtual display region are displayed within a radial graph, and the first virtual display region is positioned closer to a central portion of the radial graph than the second virtual display region is.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS This application claims priority benefit of the United States Provisional Patent Application titled, “ARTIFICIAL INTELLIGENCE-BASED TECHNIQUES FOR COLLABORATION IN VIRTUAL ENVIRONMENTS,” filed on Sep. 2, 2020 and having Ser. No. 63/073,883. The subject matter of this related application is hereby incorporated herein by reference. BACKGROUND Field of the Various Embodiments The various embodiments relate generally to computer science and computer architecture and, more specifically, to artificial intelligence-based techniques for generating designs in virtual environments. Description of the Related Art Generative design is a computer-implemented process of design exploration, in which potential design solutions for a given design problem are generated automatically via one or more software applications. In practice, generative design is an iterative process that typically generates numerous potential designs that satisfy the parameters of the given design problem, such as boundary conditions, loading, geometric limits, and other constraints. A designer then sorts through the generated designs and attempts to improve on the best potential designs. In this regard, the designer typically modifies input values for the design problem parameters with values that are likely to result in additional design solutions, where various targeted features of the additional design solutions align more closely with a desired result. Generally, generative design enables designers to explore a much larger number of design solutions than is possible using traditional 3D modeling processes. Generative design also can generate design solutions that are difficult or non-obvious for humans to create using traditional 3D modeling processes. One drawback of generative design is that a designer cannot provide inputs to generative design applications that directly cause particular features to be included in the design solutions generated by the generative design software. Instead, the designer can only modify the problem parameters in a way that indirectly results in particular features being included in the design solutions generated by the generative software. Accordingly, to generate designs that include a given desired feature, an iterative process has to be followed that can be time-consuming and frustrating for designers. Among other things, a designer has to develop an understanding of how changes to certain problem parameters can produce certain desired features, which is something that can be achieved only through repeated iterations of the generative design process. Another drawback of generative design is that generative design software operates in a way that prevents designers from staying engaged in the creative process. In this regard, with conventional approaches, a designer inputs certain problem parameters into the software and then has to wait for several minutes or hours before the software generates the potential design solutions. Aside from being inefficient, the continual and protracted interruptions associated with this type of workflow are quite disruptive to the creative process. Further, this workflow can cause the designer to forget the context for the selection of a particular set of problem parameters, which can prevent the designer from following a coherent path in the exploration of possible designs that is enabled by generative design. Without such context, each iteration of the generative design process engaged in by the designer can be disconnected from preceding iterations. As a result, the resulting designs from each iteration do not build on each other coherently, and the modifications for each iteration can be somewhat random. As the foregoing illustrates, what is needed in the art are more effective ways to generate design solutions via the generative design process. SUMMARY One embodiment of the present sets forth a technique for generating designs that includes: causing one or more candidate designs to be displayed within a virtual-reality (VR) environment; receiving a user input associated with a first candidate design included in the one or more candidate designs via the VR environment; generating a modified design based at least on the user input and the first candidate design; and generating a plurality of output designs via a generative design process based on the modified design. At least one technical advantage of the disclosed techniques relative to the prior art is that the disclosed techniques enable a user to more easily and intuitively modify the desired features of designs generated by generative design software. For example, with the disclosed techniques, a user can directly manipulate specific portions of an object mesh and/or combine multiple designs that include one or more desired features. Accordingly, with the disclosed techniques, a user is not forced to indirectly cause a generative design system to generate a design with a desired fe