Two papers were accepted at NeurIPS 2026: “Learning When to Think: Adaptive Internal Computation for Reinforcement Learning” in the main conference, and our position paper “Neurosymbolic AI is a Strong Technical Foundation for Trustworthy, Deployable AI by Design”, with collaborators at Carnegie Mellon University, in the position paper track.
We are presenting three papers at IROS 2026: "From Local Corrections to Generalized Skills: Improving Neuro-Symbolic Policies with MEMO" in the main conference, “TRACT: Bimanual Cable Insertion via Tactile Tension Regulation” at the Touch-to-Action workshop, and “Learning Object-Centric Barrier Functions for Contact-Rich Interaction in Clutter” at the Learning and Formal Methods for Robotics workshop.
Hyeonbeen Lee, Sirui Zhan, and Haixin Jin join TEA Lab as PhD students this fall. Welcome!
Our work "Learning When to Stop: Prefix-Optimal Dynamic Diffusion Policies for Continuous Control" was presented at the “Reinforcement Learning Conference (RLC 2026)” in Montréal, where it learns a prefix value function at every denoising step to give diffusion policies a parameter-free stopping criterion at test time.
Our work "Influence-Guided Agent Adaptation for Dynamic Human–AI Collaboration" was presented at the “Continual Reinforcement Learning Workshop @ RLC 2026” in Montréal.
Our work "From Local Corrections to Generalized Skills: Improving Neuro-Symbolic Policies with MEMO", in collaboration with Collab at Virginia Tech, was accepted at “IROS 2026”. MEMO builds a retrieval-augmented skillbook from natural-language corrections, clustering feedback across users and tasks into generalized skill templates that a neuro-symbolic policy can retrieve at run time.
TEA Lab starts at Virginia Tech!
We received the best-paper award at the “Generative Models x HRI Workshop @ RSS 2025” for our work on “Modeling Latent Partner Strategies for Adaptive Zero-Shot Human-Agent Collaboration”
We received one of two best-paper awards at the “Workshop on Nonverbal Cues for Human-Robot Cooperative Intelligence at ICRA 2025” for our work on “Let Me Help You! Neuro-Symbolic Short-Context Action Anticipation”