Oscar Mañas

Hola! I’m Oscar Mañas, a Research Scientist at Meta in Zurich, working on multimodal and embodied AI. I received my PhD from Mila and Université de Montréal, where I was advised by Prof. Aishwarya Agrawal. I’ve also been a Visiting Researcher at Meta FAIR and a Research Intern at Element AI, and my work has been published at CVPR, ICML, ICCV, AAAI, EACL, and TMLR. See my CV for more details.

We experience the world through many channels: we take it in through sight and touch, act on it with our bodies, and use language both ways. Each is a different view of the same underlying reality. I build multimodal models that learn from these views to understand the world, generate it, and act in it. My work has followed that path: from vision-language models and image and video generation, along with the reward models and metrics that evaluate and steer them, to video world models, and now to embodied and physical AI with vision-language-action models (VLAs) and world-action models (WAMs). Once actions are just another modality, what a model learns from watching the world can carry over to acting in it, and from there to improving itself through its own experience.