Max Robotics — Global Service Robot Directory + US Certification

About

Wilson Yan is a researcher working at the intersection of robot learning, computer vision, and generative modeling, with particular expertise in deformable object manipulation and video generation. His most recognized contribution, "Learning to Manipulate Deformable Objects without Demonstrations" (2020, 164 citations), introduced a model-free visual reinforcement learning framework that dramatically improves sample efficiency through an innovative iterative pick-place action space — enabling robots to handle challenging deformable objects without costly human demonstrations. Complementing this, his work on contrastive estimation for learning predictive representations (75 citations) advanced model-based approaches to the same problem by jointly optimizing visual representations and dynamics models. On the generative modeling front, Yan's VideoGPT (2021, 144 citations) has become a widely referenced architecture for scalable video generation, elegantly combining VQ-VAE discrete latent representations with transformer-based autoregressive modeling. More recently, his Language Quantized AutoEncoders work (2023) pushes toward unsupervised text-image alignment, addressing the visual limitations of large language models. Collectively, Yan's research has accumulated over 400 citations, reflecting meaningful impact across robotics and generative AI communities and establishing him as a versatile contributor bridging physical manipulation and multimodal learning.