FLAIR · Research

Learning, generation,
and physical intelligence.

We develop foundation models and learning algorithms for intelligent physical systems, alongside mathematical foundations for generative modeling, inference, and control.

Physical AI is a major current direction. Our work in sampling, optimal transport, and stochastic control addresses fundamental questions in inference and decision-making.

01 / Research theme

Physical AI & robotics

How can robots learn reusable representations and skills that generalize across tasks?

We develop robot foundation models, world models, and learning algorithms for intelligent physical systems. Our research in agentic robotics explores how memory, reasoning, and exploration can support sustained autonomous behavior, alongside our work on compositional skills and planning.

Current directions

  • Robot foundation models and vision-language-action learning
  • World models and physical reasoning
  • Reinforcement learning and post-training for robot policies
  • Compositional skills and long-horizon autonomy
  • Agentic robotics: memory, reasoning, exploration, and long-horizon planning

Selected contributions

Physical AI & robotics

Compositional Diffusion with Guided Search for Long-Horizon Planning

Utkarsh A. Mishra, David He, Yongxin Chen, and Danfei Xu

Guided search within compositional diffusion connects local models into coherent long-horizon plans.

CDGS · ICLR 2026

Physical AI & robotics

Generative Skill Chaining: Long-Horizon Skill Planning with Diffusion Models

Utkarsh A. Mishra, Shangjie Xue, Yongxin Chen, and Danfei Xu

Composing skill-level diffusion models generates long-horizon manipulation plans for unseen task combinations.

GSC · CoRL 2023

02 / Research theme

Generative models & learning

How can we make generative models more efficient, adaptable, and mathematically understood?

Generative modeling is an independent research focus and a source of tools for physical intelligence. We study both the underlying algorithms and their practical behavior.

Current directions

  • Diffusion and flow models
  • Reinforcement learning for generative models
  • Efficient inference and sampling
  • Mathematical principles of generative learning

Selected contributions

Generative models & learning

Scaling Reinforcement Learning for Diffusion Models via Velocity Matching

Jaemoo Choi, Wei Guo, Yuchen Zhu, Arash Vahdat, Molei Tao, Julius Berner, and Yongxin Chen

Reward-based velocity matching adapts diffusion models through updates to their velocity fields.

RVM · Preprint · 2026

Generative models & learning

Fast Sampling of Diffusion Models with Exponential Integrator

Qinsheng Zhang and Yongxin Chen

Exponential integrators exploit the structure of diffusion dynamics to reduce sampling steps.

DEIS · ICLR 2023

03 / Research theme

Control & mathematical foundations

How can we reason about distributions, dynamics, and decisions under uncertainty?

Our foundational work develops theory and algorithms with applications across control, inference, and machine learning. Optimal transport and sampling remain research goals in their own right.

Current directions

  • Stochastic control and probabilistic safety
  • Optimal transport and Schrödinger bridges
  • Markov chain Monte Carlo and sampling complexity
  • Optimization and decision-making under uncertainty

Selected contributions

Control & mathematical foundations

Probabilistic Reachability Analysis of Stochastic Control Systems

Saber Jafarpour, Zishun Liu, and Yongxin Chen

Probabilistic bounds connect deterministic reachability tools with stochastic system analysis.

Probabilistic reachability · IEEE Transactions on Automatic Control · 2025

Control & mathematical foundations

Stochastic Control Liaisons: Richard Sinkhorn Meets Gaspard Monge on a Schrödinger Bridge

Yongxin Chen, Tryphon T. Georgiou, and Michele Pavon

A survey connecting stochastic control, optimal transport, Schrödinger bridges, and Sinkhorn algorithms.

Schrödinger bridges · SIAM Review · 2021

Explore further work on Schrödinger bridges, covariance control, optimal transport, and MCMC in the selected publications.