Haoyu Ma

About

Haoyu Ma is an emerging researcher working at the intersection of reinforcement learning, world models, and unsupervised representation learning. His work focuses on developing more generalizable and data-efficient approaches to model-based reinforcement learning (MBRL), with a particular emphasis on leveraging large-scale, real-world video data for agent pre-training. His most notable contribution, "Pre-training Contextualized World Models with In-the-wild Videos for Reinforcement Learning" (2023), addresses a critical limitation in the field — the tendency of existing pre-training methods to rely on narrow, domain-specific or simulated datasets. By harnessing the richness and diversity of in-the-wild video data, Ma's approach pushes world model pre-training toward greater real-world applicability, a meaningful step in bridging the gap between controlled benchmarks and practical deployment. Though early in his research career with 3 citations on this work, his focus on unsupervised pre-training for sequential decision-making places him within one of the most rapidly evolving areas of modern AI. Students and researchers interested in sample-efficient reinforcement learning and scalable agent learning will find his contributions a compelling and timely read.

Research Focus

Key Achievements

Top Papers

1. 1
   Pre-training Contextualized World Models with In-the-wild Videos for Reinforcement Learning

   3 citations · 2023

Key Collaborators

Contact & Links