2022
Stanford University
PhD, Aeronautics & Astronautics
Multi-agent reinforcement learning at Stanford Intelligent Systems Laboratory.
Multi-agent intelligence · Stanford
Teaching intelligent systems to cooperate.
I’m Sheng Li, a Stanford Aero/Astro PhD exploring the structure of multi-agent cooperation—how agents communicate, coordinate, and make safer decisions together.
01
About
Structure before scale
Intelligence is more powerful when it knows how to work together.
At Stanford Intelligent Systems Laboratory, I studied cooperation and coordination from a structural perspective: learning who should communicate, what should be said, and how a group can act as more than a collection of individuals.
My earlier work applied reinforcement learning to aircraft collision avoidance and urban air mobility—domains where coordination is not just elegant, but essential for safety.
01
Cooperation
Shared behavior
02
Communication
Emergent language
03
Coordination
Dynamic structure
04
Safety
Autonomous airspace
2019
Stanford University
MS, Aeronautics & Astronautics
Autonomous systems, decision making, and airspace safety.
2017
Michigan × SJTU
Dual bachelor’s degrees
Aerospace Engineering and Mechanical Engineering.
02
Selected research
Systems that think together
Selected work.
From learned language to dense airspace, each project asks how local decisions can create intelligent collective behavior.
01
Emergent discrete messages for cooperative learning
Agents learn a compact, interpretable vocabulary through a broadcast-and-listen mechanism—with a path for humans to join the conversation.
Open project ↗
02
Deep implicit coordination graphs
A dynamic graph structure that learns which agent interactions matter, balancing centralized reasoning with decentralized action.
Open project ↗
03
Best paper
Fleet management at city scale
Simulation-driven analysis of fleet, infrastructure, and traffic policies for scalable on-demand urban air mobility.
Open project ↗
04
Collision avoidance in dense airspace
Deep reinforcement learning corrections help traditional avoidance systems maintain safety while operating more efficiently at high traffic density.
Open project ↗More experiments in autonomous driving, uncertainty, perception, and distributed learning.
Enter the research index →
03
Publications
Peer-reviewed ideas
In print.
2022
Learning Emergent Discrete Message Communication for Cooperative Reinforcement Learning
ICRA
↗
2021
Deep Implicit Coordination Graphs for Multi-agent Reinforcement Learning
AAMAS
↗
2020
Towards a Distributed Framework for Multi-Agent Reinforcement Learning Research
HPEC
↗
2020
Analysis of Fleet Management and Infrastructure Constraints in On-Demand Urban Air Mobility Operations
AIAA · Best paper
↗
2019
Optimizing Collision Avoidance in Dense Airspace Using Deep Reinforcement Learning
ATM Seminar
↗
04
Beyond the lab
Light · place · patience
Looking closely is another kind of research.
Away from algorithms, I photograph landscapes and cities—waiting for scale, motion, and light to briefly align.
Follow the visual journal ↗
Open channel
Let’s coordinate.
Research, intelligent systems, ambitious ideas—or a great place to take a photograph.
lisheng@stanford.edu ↗