Agrim Gupta
Agrim Gupta is a computer vision and robotics researcher whose work spans human-motion prediction, object segmentation, robot learning and video generation. He began in engineering and research roles at Microsoft, Stanford and Meta, then studied for a Stanford PhD under Fei-Fei Li from 2019 to 2024; he holds a bachelor's degree from BITS Pilani. His 2017 work on neural style transfer preceded Social GAN, LVIS and robot-learning systems such as MetaMorph, VIMA and RoboCat. He coauthored VideoPoet, which received an ICML 2024 Best Paper award, then contributed to Google DeepMind's Genie 3 and coauthored MALT Diffusion in 2025. He is a research scientist at Meta, a role he began in December 2025.
Timeline
-
2016–2019
Early roles and style transfer
In 2016, he interns at Microsoft and works as a research assistant at Stanford; from 2017 to 2019, he is a research engineer at Meta AI Research. His 2017 ICCV paper on neural style transfer uses a recurrent network with a temporal consistency loss to overcome instability in video.
-
2018
Social GAN predicts motion
In 2018, he coauthors Social GAN, a generative model that predicts socially plausible future human trajectories.
-
2019
LVIS and instance segmentation
In 2019, he coauthors LVIS, a dataset for large-vocabulary instance segmentation covering over 1,000 object categories.
-
2019–2024
Doctoral work and robot models
From 2019 to 2024, he is a doctoral student at Stanford advised by Fei-Fei Li; his bachelor's degree is from BITS Pilani. In 2022, he coauthors MetaMorph, a transformer-based universal controller for modular robots. In 2023, he coauthors VIMA, which uses multimodal prompts for general robot manipulation.
-
2023
RoboCat learns across robots
He co-leads RoboCat, a 2023 robot manipulation agent that generalizes to new tasks and robots and uses its own generated data to improve through further training.
-
2024
VideoPoet wins an ICML award
He coauthors VideoPoet, a language model that synthesizes high-quality video from multimodal inputs. The paper receives ICML's 2024 Best Paper award.
-
2025
Genie 3 and MALT Diffusion
At Google DeepMind, he contributes to Genie 3 in 2025, a world model that generates interactive worlds users can explore in real time. He coauthors MALT Diffusion, a 2025 diffusion model that uses memory-augmented latent transformers to generate long videos.
-
Since December 2025
Research scientist at Meta
He is a research scientist at Meta after joining in December 2025.
Career
- Since 2025MetaResearch Scientist
- 2025Google DeepMind
- 2017–2019FAIR (Meta)Research Engineer
- 2016MicrosoftIntern
- 2016Stanford UniversityResearch Assistant
Education
- 2024Stanford UniversityPhD
- Birla Institute of Technology and Science, PilaniBachelor's
Sources and credits
Photo: We have no published photo of Agrim Gupta yet. Send us a free-licensed photo.
- Find Agrim Gupta's email & phone number - Research Scientist at Meta, San Francisco Bay Area | SignalHire
- Characterizing and Improving Stability in Neural Style Transfer
- [1803.10892] Social GAN: Socially Acceptable Trajectories with Generative Adversarial Networks
- [1908.03195] LVIS: A Dataset for Large Vocabulary Instance Segmentation
- Agrim Gupta
- [2203.11931] MetaMorph: Learning Universal Controllers with Transformers
- [2210.03094] VIMA: General Robot Manipulation with Multimodal Prompts
- [2306.11706] RoboCat: A Self-Improving Generalist Agent for Robotic Manipulation
- VideoPoet: A Large Language Model for Zero-Shot Video Generation
- ICML 2024 Awards
- Genie 3: A new frontier for world models
- [2502.12632] MALT Diffusion: Memory-Augmented Latent Transformers for Any-Length Video Generation


