(Cannon Beach, Oregon)

Haonan Yu


Google scholar / GitHub / Blog / LinkedIn

Hi there👋 I am a research scientist at Horizon Robotics. Previously, I worked at Baidu Research and Facebook AI Research.

My early research involved various topics such as object segmentation, language grounding in video, video captioning , embodied language understanding, and the lottery ticket hypothesis. Since joining Horizon Robotics in 2019, I've been researching novel RL algorithms (hierarchical RL, safe/constrained RL, and exploration, etc) and building learning-based robotic applications.

Over the course of my career, as the first author I have published papers in top conferences such as ACL, CVPR, NeurIPS, ICLR, CoRL, CVPR, and AAAI, covering a wide range of topics including efficient RL, safe RL, robotics, language grounding, embodied AI, multimodal learning, image/video segmentation, etc. I strongly believe that realizing AGI necessitates the integration of diverse disciplines, and thus I have been actively engaging with various AI domains that attract my interests. My endeavors have been recognized with distinction, particularly marked by the receipt of the Best Paper Award at ACL 2013 for my work on grounded language learning from video.

I love writing code and building things. I am a co-author and an active maintainer of the RL framework ALF that supports most RL research at Horizon.


My current passion is to combine all my past experience on language grounding, multimodal learning, and learning-based robotic control, to build an agile robot, with mobile manipulation capabilities, that follows natural language commands in daily indoor scenarios.


Education

Ph.D. of Electrical and Computer Engineering (ECE), Purdue University
Bachelor of Computer Science (CS), Peking University


Selected publications


Softwares

ALF - Agent Learning Framework (ALF) is a PyTorch reinforcement learning framework emphasizing on the flexibility and easiness of implementing complex algorithms involving many different components.
TAAC - Temporally abstract actor-critic algorithm for efficient continuous control.
XWorld - A simulator package for RL research with language task support.
FLARE - A reinforcement learning framework for training embodied agents with PyTorch.


Hobbies

I like playing video games, especially ARPG. So far my favorite games are souls-like, by FromSoftware.