COVER STORY - AGE OF AI
Bernini is an experimental generative AI proof-of-concept trained of the world’s largest training dataset of 3D shapes. The lessons Autodesk’s AI researchers learned when developing Bernini have helped the team immensely as they pursue additional generative models.”
FUTURE-PROOFING DESIGN AND MAKE While Bernini was developed purely as proof-of-concept with no plan for commercialisation, the model offers a promising outlook for generative AI within the future of product design. According to Autodesk’s 2024 State of Design & Make report, 78% of business leaders believe AI will enhance their industry and 79% agree that AI will make their industry more creative. “Project Bernini points to a future
when professionals will sketch or describe an object using natural language and the technology quickly generates geometry that a designer could import and begin to edit and adapt for their needs,” Green says. Through Bernini, Autodesk’s
engineers focused on generating functional 3D structures that could work in the real world and serve the purpose that the designer has in mind.
GENERATIVE CAPABILITIES So how does the model work? “Project Bernini accepts a variety of unput modes including natural language text prompts, 2D sketches, point clouds, voxels and more to generate 3D shapes,” Green explains. “We continue to research new input modes and ways to enhance precision and more.” To best illustrate the model’s
capabilities, the example of a water pitcher is given. While other 3D generative models might produce shapes that look like the pitcher with textures that improve their superficial appearance in a specific lighting environment, the Bernini model generates shape and texture separately and does not confuse or meld those variables. Therefore, water pitchers generated by Bernini would be hollow in the middle and could actually hold water, as any real-world pitcher would need to do. Autodesk’s AI Lab and Research teams trained the Bernini model on
Demonstrating Project Bernini
a dataset of 10 million diverse 3D shapes, made up of publicly available data, a mixture of CAD objets and organic shapes. “The research team learned how
to collect and curate the world’s largest training dataset of 3D shapes, train the model, balance compute resources, and evolve the inference results,” Green continues. “These are all valuable lessons as they continue to focus on topical research efforts in generative AI.”
LOOKING AHEAD While Project Bernini is exciting and at the leading edge of generative AI, Autodesk states its models will become increasingly useful when trained on larger, higher-quality professional datasets. But how disruptive could generative AI be to the product design and engineering process as a whole? “Initially, we expect to see many
time-consuming, tedious tasks automated to such a degree that much of the non-creative and less valuable tedium designers face today with these tools is streamlined or eliminated altogether,” Green predicts. “The
efficiency gains will represent a type of disruption. However, the greater disruptions will come when designers and engineers are able to partner more fully with the software to generate unique designs and systems not only faster but in unique ways.” In fact, Project Bernini is just one
aspect of a much larger AI strategy that Autodesk is deploying, focused on delivering intelligent assistance and generative capabilities across its entire cross-industry product portfolio. “Today, the design of a physical
product is the result of often many thousands of small steps within existing design software,” Green adds. “For example, a cylinder might be extruded and then narrowed. Each step represents a software command. Imagine a future that allows a product to be created from an AI- generated series of commands where these commands are automatically inferred purely from the designer’s description. There will be disruptions, but the optimistic view is that pairing human ingenuity with generative AI- enabled systems will create capacity for this partnership to achieve more.”
www.engineerlive.com 11
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52
