Student Speakers - Yuejiang Liu & Wenlong Huang, "Robot Learning Without Action Chunking” & “Generalization through Task Representations with Foundation Models”
Date: May 23, 2025 @ 3:00 - 4:00PM | Location: Gates B03
The seminar is open Stanford students, staff, and affiliates.
Title: Generalization through Task Representations with Foundation Models
Speaker: Wenlong Huang
Time: Friday May 23th, 3:00-3:30PM
Abstract: Building robots that can operate autonomously in unstructured environments by following arbitrary natural language commands has long been the north star in robotic manipulation. While there has been tremendous progress in learning visuomotor policies that exhibit promising signs for open-world deployment, generalization to unseen tasks or motions largely remains unattainable or out of scope. In this talk, I will discuss how deliberate choices of task representations enable such zero-shot generalization at the task level, despite given no task-specific demonstrations. Notably, I will discuss our years-long investigations into extracting task representations from off-the-shelf foundation models; I will discuss its evolution from a language-only representation to 4D space-time domain and their applications to model-based planning, affordance learning, and visuomotor policy learning. At the end of the talk, I will present an alternative view for scaling towards robotic intelligence: by leveraging foundation models to provide task-specific knowledge in the form of task representations, robotic data scaling can focus on learning from task-agnostic interactions with a world modeling objective, such that collectively this enables robots that not only understand the world as humans do but can also act within it with purpose and generality.
Bio: Wenlong Huang is a PhD candidate in Computer Science at Stanford University, advised by Professor Fei-Fei Li. He received his B.A. in Computer Science from UC Berkeley, where he was advised by Professor Deepak Pathak, Dr. Igor Mordatch, and Professor Pieter Abbeel. He studies the intersection between robotic manipulation, foundation models, and 3D computer vision. His works have won the Outstanding Paper Award in Robot Learning at ICRA 2023, the Best Paper Award at the CoRL 2024 LEAP Workshop, and the Best Paper Finalist at ICRA 2025. He received Stanford School of Engineering Fellowship and was selected as a finalist for the NVIDIA Graduate Fellowship and the Citadel GQS Fellowship.
Title: Robot Learning Without Action Chunking
Speaker: Yuejiang Liu
Time: Friday May 23th, 3:30-4:00PM
Abstract: Recent advances in robot learning have mirrored the progress of large language models in many ways—yet, one key distinction remains: action chunking. In this talk, I will begin with an analysis of action chunking, highlighting its inherent tradeoff between long-term consistency and short-term reactivity. I will then introduce two methods to address this tradeoff: (i) Bidirectional Decoding: an inference algorithm that jointly optimizes consistency and reactivity using additional compute at test time; (ii) Past-Token Prediction: an auxiliary training objective that encourages diffusion policies to capture temporal dependencies in long-context observations. Together, these methods offer a promising path toward memory-aware robot policies without action chunking.
Bio: Yuejiang Liu is a Postdoc in Computer Science at Stanford University, advised by Chelsea Finn. His research focuses on learning algorithms for robust and adaptive robots in dynamic interactive environments. Recently, he has been working on improving scalability through data curation and test-time inference.
Please visit https://stanfordasl.github.io/robotics_seminar/ for this quarter’s lineup of speakers. Although we encourage live in-person attendance, recordings of talks will be posted also.