Shubham Goel



CV | Google Scholar

Hello! I'm a research scientist working on generative video and reward modeling. Most recently, at SpaceXSI (formerly xAI), I led evaluation for Grok Imagine and shipped Video 1.0 and Video 1.5.

Before that, I headed the research team at Avataar, working on LLM distillation, diffusion models, and 3D reconstruction.

I'm currently exploring my next role in robotics, LLMs/VLMs, and world models. Feel free to reach out!

I received my PhD from UC Berkeley under the guidance of Jitendra Malik and Angjoo Kanazawa, and my undergrad from IIT Bombay. I also spent some time at Facebook AI Research during grad school, and at TU Darmstadt and IST Austria during undergrad.

Publications

Spatial Cognition from Egocentric Video: Out of Sight, Not Out of Mind
The More You See in 2D, the More You Perceive in 3D
Humans in 4D: Reconstructing and Tracking Humans with Transformers
Differentiable Stereopsis: Meshes from multiple views using differentiable rendering
Amazon Berkeley Objects (ABO) Dataset
Shape and Viewpoint without Keypoints
Boolean Functional Synthesis: Hardness and Practical Algorithms‡
What's hard about Boolean Functional Synthesis?‡
Computing Scores of Forwarding Schemes in Switched Networks with Probabilistic Faults‡

* Equal contribution. † Equal advising. ‡ Authors sorted alphabetically by last name.

Connections

github X (Twitter) linkedin

Many thanks to Shumo Chu for the site inspiration.