Brandon Houghton

Papers

Total Citations

H-Index

About

Brandon Houghton is a leading researcher in the intersection of artificial intelligence, robotics, and sequential decision-making, with a particular focus on leveraging large-scale, unlabeled data to train generalist agents. His most influential contribution is the Video PreTraining (VPT) framework, introduced in his highly cited 2022 paper (50+ citations), which pioneered a method for pretraining agents to act by simply watching unlabeled online videos. This breakthrough demonstrated that models can acquire broad, general capabilities for domains like video games, robotics, and computer use without expensive human demonstrations or handcrafted reward functions. By harnessing noisy, internet-scale video datasets, Houghton’s work challenges traditional reinforcement learning paradigms, offering a scalable path toward more adaptable and autonomous systems. His research has profound implications for reducing the data bottleneck in embodied AI, making it possible to train agents that learn from the vast repository of human behavior captured in everyday video. Houghton’s innovative approach has positioned him at the forefront of efforts to build foundation models for action, inspiring a new wave of research into pretraining for sequential decision-making.