01
Reinforcement Learning
Agents that learn efficiently, reason over long horizons, and improve through interaction with complex environments.
RLAgentsDecision Making
BEIHANG AI LAB · BEIJING
Intelligence, engineered to scale.
We study learning, reasoning, agents, and optimization — building AI systems that can learn from experience, reason about difficult problems, and act effectively in the world.
BEIHANG / AI SYSTEMS● ACTIVE
BEIHANG UNIVERSITYSCROLL TO DISCOVER ↓
01 / ABOUT
We work at the boundary of learning, reasoning, and optimization.
Beihang AI Lab is an open research group focused on capable AI systems and the principles behind them. Our work ranges from reinforcement learning and foundation models to embodied intelligence and large-scale combinatorial optimization.
01
Simple ideas
We prefer clear mechanisms over unnecessary complexity.
02
Strong evidence
We value careful experiments, reproducibility, and honest analysis.
03
Systems that scale
We care about methods that remain useful as models, data, and problem sizes grow.
02 / RESEARCH
Research directions
Four connected directions, one goal: building more capable and useful intelligent systems.
02
Foundation Models & Alignment
Post-training, alignment, reasoning, and reliable adaptation for increasingly capable foundation models.
LLMAlignmentPost-training
03
Embodied Intelligence
World models and embodied agents that connect perception, memory, planning, and action in the physical world.
Embodied AIWorld ModelsPlanning
04
Combinatorial Optimization
Learning-based and training-free methods for large-scale routing and other structured optimization problems.
RoutingOptimizationNeural CO
03 / SELECTED RESEARCH
What we are building
AUTOMATED ALGORITHM DESIGN
PUBLICDG2AD
Directed graph-guided automated algorithm design with large language models, targeting modular and evolvable optimization procedures.
Code & projectLARGE-SCALE ROUTING
RESEARCHJust Initialize
A training-free component for large-scale routing optimization that discovers refinement-friendly initializations before downstream solving.
Project page soon04 / NEWS
Latest from the lab
Research releases, open-source updates, and lab milestones.
01
Beihang AI Lab launches a new public home for our research, code, and collaborations.
02
Just Initialize is released as a training-free framework for large-scale routing optimization.
03
DG2AD source code is released publicly through the BeihangAILab GitHub organization.
05 / PEOPLE
Small team.
Big questions.
We are building an open and ambitious research community at Beihang University. We welcome collaborators interested in reinforcement learning, foundation models, embodied intelligence, and optimization.
06 / CONNECT