_version_ 1866916850049220608
author Foundation AI Team
Bhat, Kiran
Khanna, Nishchaie
Channa, Karun
Zhou, Tinghui
Zhu, Yiheng
Sun, Xiaoxia
Shang, Charles
Sudarshan, Anirudh
Chu, Maurice
Li, Daiqing
Deng, Kangle
Fauconnier, Jean-Philippe
Verhulsdonck, Tijmen
Agrawala, Maneesh
Fatahalian, Kayvon
Weiss, Alexander
Reiser, Christian
Chirravuri, Ravi Kiran
Kandur, Ravali
Pelaez, Alejandro
Garg, Akash
Palleschi, Michael
Wang, Jessica
Litz, Skylar
Liu, Leon
Li, Anying
Harmon, David
Liu, Derek
Feng, Liangjun
Goupil, Denis
Kuczynski, Lukas
Yoon, Jihyun
Marri, Naveen
Zhuang, Peiye
Zhang, Yinan
Yin, Brian
Jiang, Haomiao
van Workum, Marcel
Lane, Thomas
Erickson, Bryce
Pathare, Salil
Price, Kyle
Han, Steve
Wang, Yiqing
Singh, Anupam
Baszucki, David
author_facet Foundation AI Team
Bhat, Kiran
Khanna, Nishchaie
Channa, Karun
Zhou, Tinghui
Zhu, Yiheng
Sun, Xiaoxia
Shang, Charles
Sudarshan, Anirudh
Chu, Maurice
Li, Daiqing
Deng, Kangle
Fauconnier, Jean-Philippe
Verhulsdonck, Tijmen
Agrawala, Maneesh
Fatahalian, Kayvon
Weiss, Alexander
Reiser, Christian
Chirravuri, Ravi Kiran
Kandur, Ravali
Pelaez, Alejandro
Garg, Akash
Palleschi, Michael
Wang, Jessica
Litz, Skylar
Liu, Leon
Li, Anying
Harmon, David
Liu, Derek
Feng, Liangjun
Goupil, Denis
Kuczynski, Lukas
Yoon, Jihyun
Marri, Naveen
Zhuang, Peiye
Zhang, Yinan
Yin, Brian
Jiang, Haomiao
van Workum, Marcel
Lane, Thomas
Erickson, Bryce
Pathare, Salil
Price, Kyle
Han, Steve
Wang, Yiqing
Singh, Anupam
Baszucki, David
contents Foundation models trained on vast amounts of data have demonstrated remarkable reasoning and generation capabilities in the domains of text, images, audio and video. Our goal at Roblox is to build such a foundation model for 3D intelligence, a model that can support developers in producing all aspects of a Roblox experience, from generating 3D objects and scenes to rigging characters for animation to producing programmatic scripts describing object behaviors. We discuss three key design requirements for such a 3D foundation model and then present our first step towards building such a model. We expect that 3D geometric shapes will be a core data type and describe our solution for 3D shape tokenizer. We show how our tokenization scheme can be used in applications for text-to-shape generation, shape-to-text generation and text-to-scene generation. We demonstrate how these applications can collaborate with existing large language models (LLMs) to perform scene analysis and reasoning. We conclude with a discussion outlining our path to building a fully unified foundation model for 3D intelligence.
format Preprint
id arxiv_https___arxiv_org_abs_2503_15475
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Cube: A Roblox View of 3D Intelligence
Foundation AI Team
Bhat, Kiran
Khanna, Nishchaie
Channa, Karun
Zhou, Tinghui
Zhu, Yiheng
Sun, Xiaoxia
Shang, Charles
Sudarshan, Anirudh
Chu, Maurice
Li, Daiqing
Deng, Kangle
Fauconnier, Jean-Philippe
Verhulsdonck, Tijmen
Agrawala, Maneesh
Fatahalian, Kayvon
Weiss, Alexander
Reiser, Christian
Chirravuri, Ravi Kiran
Kandur, Ravali
Pelaez, Alejandro
Garg, Akash
Palleschi, Michael
Wang, Jessica
Litz, Skylar
Liu, Leon
Li, Anying
Harmon, David
Liu, Derek
Feng, Liangjun
Goupil, Denis
Kuczynski, Lukas
Yoon, Jihyun
Marri, Naveen
Zhuang, Peiye
Zhang, Yinan
Yin, Brian
Jiang, Haomiao
van Workum, Marcel
Lane, Thomas
Erickson, Bryce
Pathare, Salil
Price, Kyle
Han, Steve
Wang, Yiqing
Singh, Anupam
Baszucki, David
Computer Vision and Pattern Recognition
Foundation models trained on vast amounts of data have demonstrated remarkable reasoning and generation capabilities in the domains of text, images, audio and video. Our goal at Roblox is to build such a foundation model for 3D intelligence, a model that can support developers in producing all aspects of a Roblox experience, from generating 3D objects and scenes to rigging characters for animation to producing programmatic scripts describing object behaviors. We discuss three key design requirements for such a 3D foundation model and then present our first step towards building such a model. We expect that 3D geometric shapes will be a core data type and describe our solution for 3D shape tokenizer. We show how our tokenization scheme can be used in applications for text-to-shape generation, shape-to-text generation and text-to-scene generation. We demonstrate how these applications can collaborate with existing large language models (LLMs) to perform scene analysis and reasoning. We conclude with a discussion outlining our path to building a fully unified foundation model for 3D intelligence.
title Cube: A Roblox View of 3D Intelligence
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2503.15475