Abstract: A method and system provide the ability to optimize a structural engineering design. A dataset is synthesized by acquiring a structural skeleton design of an entire building. The skeleton defines locations and connectivities of bars that represent columns or beams. The skeleton design is represented as a structural graph with each bar represented as a graph node and edges connecting graph nodes. Structural simulation results are computed for the synthetic dataset based on the structural graph, various loads, and a structural analysis. A simulation model and a size optimization model are trained based on the structural simulation results with the size optimization model determining cross-section sizes for the bars to satisfy a building mass objective, building constraints, and output from the simulation model. The structural engineering design is output from the size optimization model.

Abstract: An agent engine allocates a collection of agents to scan the surface of an object model. Each agent operates autonomously and implements particular behaviors based on the actions of nearby agents. Accordingly, the collection of agents exhibits swarm-like behavior. Over a sequence of time steps, the agents traverse the surface of the object model. Each agent acts to avoid other agents, thereby maintaining a relatively consistent distribution of agents across the surface of the object model over all time steps. At a given time step, the agent engine generates a slice through the object model that intersects each agent in a group of agents. The slice associated with a given time step represents a set of locations where material should be deposited to fabricate a 3D object. Based on a set of such slices, a robot engine causes a robot to fabricate the 3D object.

Abstract: A method and system provide for generating a stormwater overland flow map. Simulation inputs (ground surface data) and simulation outputs (stormwater overland flow maps) are obtained from a deluge simulation model that simulates where water will channel and accumulate on a surface. A convolutional neural network (CNN) is trained to approximate the simulation outputs of the deluge simulation model. The CNN is a sequence of CNN models that each represent a time step and each CNN model in the sequence takes CNN output from a previous CNN model as its input. The CNN output is a video output and a visual representation of stormwater overland flow over time. A new input is a first format is obtained. A collection of custom objects representing points of a grid are stored and includes z-values of points on the surface and stormwater controls. The grid is populated and then processed in/by the CNN.

Abstract: A method and system provide the ability to process a construction domain query. A natural language user query is obtained within a construction software system. The user query is pre-processed to validate the query. Text from the query is embed into search vectors for a semantic search. A data source having multiple different sections is obtained. The semantic search is performed within each of the sections and identifies semantically relevant sections. The relevant sections are consolidated into a contextual data prompt that is input into an LLM. The LLM, which is trained based on construction data, generates a response that identifies the relevant sections. The response and an identification of the relevant sections is output.

Abstract: One or more embodiments disclose techniques for generating a layout for a physical space or a building. The techniques include generating a layout based on a floorplan of a physical space, generating a plurality of spatial unit grids corresponding to a plurality of spatial units to be placed in the physical space, identifying a placement for a first spatial unit grid in the plurality of spatial unit grids within the layout by matching a corner cell in the first spatial unit grid with a given available cell in the layout, generating a score associated with the placement for the first spatial unit grid based on one or more placement parameters that define placement constraints for a first spatial unit included in the plurality of spatial units and corresponding to the first spatial unit grid, and placing the first spatial unit grid in the layout based on the score.

Abstract: In various embodiments, a parameter domain graph application generates UV-net representations of 3D CAD objects for machine learning models. In operation, the parameter domain graph application generates a graph based on a B-rep of a 3D CAD object. The parameter domain graph application discretizes a parameter domain of a parametric surface associated with the B-rep into a 2D grid. The parameter domain graph application computes at least one feature at a grid point included in the 2D grid based on the parametric surface to generate a 2D UV-grid. Based on the graph and the 2D UV-grid, the parameter domain graph application generates a UV-net representation of the 3D CAD object. Advantageously, generating UV-net representations of 3D CAD objects that are represented using B-reps enables the 3D CAD objects to be processed efficiently using neural networks.

Abstract: Methods, systems, and apparatus, including medium-encoded computer program products, for designing and manufacturing physical objects including lattice structures include, in one aspect, a method including: providing a three-dimensional model including a dual representation of a lattice structure, wherein both a shell mesh model and a solid body model of the lattice structure are producible from an additional model of the lattice structure, and beams of the lattice structure in the solid body model are hollow; performing numerical simulation using at least the shell mesh model of the dual representation to produce a current numerical assessment; modifying the additional model of the dual representation based on the current numerical assessment; repeating the performing and the modifying one or more times until the numerical simulation indicates the lattice structure satisfies at least one response requirement; and providing at least the solid body model for use in manufacturing the lattice structure.

Abstract: In various embodiments, a room-based design application automatically generates a design for a floor structure of a building. In operation, the room-based design application generates blocks representing portions of the floor structure that are delineated by walls based on a room plan included in a computer-aided design of the building. The room-based design application modifies the blocks based on a spanning length to generate partitions. Each partition represents a different portion of the floor structure that can be spanned by a structural element that has the spanning length in at least one direction. Based on the partitions, the room-based design application determines wall classifications for the walls. The room-based design application generates at least a portion of the design for the floor structure based on the wall classifications and the partitions.

Abstract: A method of modifying metadata associated with a media file is described the method comprising receiving a first media file; receiving a first metadata associated with the first media file; receiving an identifier of a second media file; and modifying the first metadata to include the identifier of the second media file.

Abstract: A method and system provide the ability to design a terrain surface. A triangular surface mesh representative of an existing surface is obtained and consists of triangles that are connected by vertices and edges. A drain intention is specified for the terrain surface through a geometry that is a point or line. The drain intention defines a drainage flow that influences a shape of the terrain surface. The mesh is modified to prevent a drain conflict between mesh triangles. A drain direction is autonomously determined for each of the mesh triangles based on the drain intention. The determination generates a drain pattern that is used to shape the terrain surface.

Abstract: Various embodiments set forth systems and techniques for generating seams for a 3D model. The techniques include generating, based on the 3D model, one or more inputs for one or more trained machine learning models; providing the one or more inputs to the one or more trained machine learning models; receiving, from the one or more trained machine learning models, seam prediction data generated based on the one or more inputs; and placing one or more predicted seams on the 3D model based on the seam prediction data.

Abstract: A method and system provide for automating drawing. A drawing of two or more entities is obtained and a resolution is determined. Based on the resolution, the drawing is quantized into a cell map. The cell map is a collection of multiple cells stored in a contiguous memory. Each of the multiple cells is quantized geometry data for and provides a smallest tangible unit of information about a corresponding entity. Each of the multiple cells is a number that represents domain specific information about the corresponding entity. The quantizing rounds off of values beyond a specified threshold to a resolution tolerance of the resolution. The cell map is utilized to automate modifications to the drawing.

Abstract: A method and system provide the ability to determine a hydrant fire flow. Inputs are obtained A critical element is identified. Based on a physical-based heuristic, a new hydrant fire flow guess is determined. The search direction is evaluated and used to maintain/override (using a heuristic method) the fire flow guess. The new guess is assigned as a hydrant demand. Network pressure and flow values are updated. The constraints are evaluated. The guess is reduced if at least one constraint has been violated and increased of all constraints have been satisfied. The new guess is evaluated for convergence and if not converged, the process repeats.

Abstract: A method and system provide for extracting an isosurface. A set of three-dimensional (3D) asymmetric sampling grids are created based on a sampling resolution and a region of interest. The set of grids cover a 3D scalar field and are tiles made up of tetrahedrons. The 3D scalar field is evaluated based on the set of 3D asymmetric sampling grids to generate a value for each tile, convert the values to cells, and assign an index to each cell. For each cell, the index is utilized to identify, in a lookup table, an enumerated tetrahedron. For each identified enumerated tetrahedron that contains an isosurface crossing, an isosurface connection is estimated between new points of edges of the identified tetrahedron. A mesh is created to connect the new points and utilized as the extracted isosurface.

Abstract: Techniques are disclosed for storing and accessing data in an information model. In some embodiments, data is stored in the information model using assets, components, and relationships, or using any other suitable transformation of the data. Each asset is a container for components. Each component stores data according to a schema that defines the organization of data in the component. Assets and components can be related to other assets and/or other components. A file can be imported into the information model by extracting data from the file, creating assets and components in the information model based on the extracted data, and wiring together the created assets and components. In addition or alternatively, an application can invoke an API to write data to the information model. An application can also invoke the API to query the information model, such as to read data from the information model.

Abstract: A voxel-based design approach enables the creating and modifying of a design model comprising a 3D grid of discrete voxels that is represented by a voxel data structure. The voxel data structure comprises voxel-level entries, each entry corresponding to a voxel based on the 3D location within the 3D grid. The voxel data structure includes a design-level entry for storing design-level performance metrics. The system updates the voxel data structure to reflect user modifications to the design model and renders a visualization of the updated design model. The system displays a per-voxel heat map for the design model for a selected performance metric based on the voxel data structure. The design system displays multiple optimized design solutions based on corresponding optimized voxel data structures. The system generates the multiple optimized design solutions based on a voxel-based optimization technique. The system also performs a voxel-based recommendation visualization technique.

Abstract: A method and system provide the ability to build a game world. A story is obtained that provides a textual narrative of a sequence of events. Plot facilities and a set of constraints are extracted from the story. Each of the plot facilities is a conceptual location where an event happens in the story. Each constraint defines a spatial relation between plot facilities. A map is generated based on the set of constraints by: generating a terrain of two dimensional (2D) polygons that is each associated with a biome type, and assigning each plot facility to a point on the terrain. The assigning complies with a maximum number of constraints and utilizes reinforcement learning (RL) to optimize positions of the points.

Abstract: A method and system provide the ability to utilize three-dimensional (3D) models to perform a predictive task. Multiple 3D models, consisting of non-Euclidean data, are obtained. Each 3D model is translated into a relational graph with nodes and edges. Each relational graph is processed using a graph neural network (GNN) that computes a node representation per node. The node representations are aggregated into a structural representation of the 3D model. Multiple different views of the 3D model are captured and passed through a convolutional neural network (CNN) to compute a view representation of each view. The view representations are aggregated into a single visual representation. The GNN and CNN are trained using a multiview contrastive training objective to maximize agreement between the structural representation and the single visual representation to form final learned representations. The final learned representation is utilized to perform the predictive task.

Abstract: Techniques are disclosed for determining users, tasks, and levels of expertise in fabrication environments. Sensors may be placed in a fabrication environment, instrumented in tools, and/or worn on users. Given labeled features extracted from sensor data, machine learning techniques are used to train models to recognize users, tasks, and levels of expertise. The trained models may then be deployed to determine the users, tasks, and levels of expertise in a fabrication environment based on features extracted from new sensor data. In turn, actions such as displaying guidance or instructional materials, or locking a user out of using certain tools or parts of the fabrication environment, may be taken.

Abstract: One embodiment of the present invention sets forth a technique for extracting data from an architectural drawing. The technique includes performing one or more operations via one or more machine learning models to extract a first image of a floorplan area from the architectural drawing. The technique also includes generating a boundary segmentation based on the first image of the floorplan area, wherein the boundary segmentation includes one or more boundary types for one or more portions of the floorplan area.