Songyang Han

Songyang Han

Research Scientist | Ph.D. | Artificial Intelligence

Sony AI

Biography

I am a research scientist at Sony AI. I got my Ph.D. degree in computer science and engineering working on artificial intelligence at the University of Connecticut. I was working on game theoretic energy management approaches in Dynamic Systems Control Lab at the University of Michigan-Shanghai Jiao Tong University Joint Institute.

My current research interests include artificial intelligence, reinforcement learning, generative AI, machine learning, and computer vision.

News

  • [2023/8] I am excited to join Sony AI as a research scientist.
  • [2023/6] Our paper “What is the Solution for State-Adversarial Multi-Agent Reinforcement Learning?” is accepted by the Frontiers4LCD Workshop at the International Conference on Machine Learning (ICML) 2023. Available on arxiv, website.
  • [2023/6] Our paper “Towards Safe Autonomy in Hybrid Traffic: Detecting Unpredictable Abnormal Behaviors of Human Drivers via Information Sharing” is accepted by ACM Transactions on Cyber-Physical Systems (TCPS).
  • [2023/5] Our paper “Robust Multi-Agent Reinforcement Learning with State Uncertainty” is accepted by Transactions on Machine Learning Research (TMLR).
  • [2023/5] I receive the CSE Predoctoral Fellowship.
  • [2023/5] I give an invited Seminar “Safe, Stable, and Robust Multi-Agent Reinforcement Learning for Connected Autonomous Vehicles” at MIT.
  • [2023/2] Our paper “Shared Information-Based Safe And Efficient Behavior Planning For Connected Autonomous Vehicles” gets the Best Paper Award in the DCAA workshop at AAAI 2023, Washington, DC. Available on arxiv.
  • [2023/1] Our paper “Uncertainty Quantification of Collaborative Detection for Self-Driving” is accepted by the 2023 IEEE International Conference on Robotics and Automation (ICRA), available on arxiv, website.
  • [2023/1] Our paper “Spatial-Temporal-Aware Safe Multi-Agent Reinforcement Learning of Connected Autonomous Vehicles in Challenging Scenarios” is accepted by the 2023 IEEE International Conference on Robotics and Automation (ICRA), available on arxiv.
  • [2022/9] Our paper “A Multi-Agent Reinforcement Learning Approach For Safe and Efficient Behavior Planning Of Connected Autonomous Vehicles” is available on arxiv, website.
  • [2022/8] I get the General Electric (GE) fellowship of excellence. The GE Fellowship for Excellence program is established to recognize excellence of current graduate students and to facilitate their completion of the Ph.D. program.
  • [2022/7] Our paper “Towards Safe Autonomy in Hybrid Traffic: The Power of Information Sharing in Detecting Abnormal Human Drivers Behaviors” is presented in the AI4TS workshop at the 31st International Joint Conference On Artificial Intelligence (IJCAI 2022).
  • [2022/7] Our paper “DeResolver: A Decentralized Negotiation and Conflict Resolution Framework for Smart City Services” is accepted by ACM Transactions on Cyber-Physical Systems. (available online).
  • [2022/5] Our paper “Stable and Efficient Shapley Value-Based Reward Reallocation for Multi-Agent Reinforcement Learning of Autonomous Vehicles” is presented on the 2022 IEEE International Conference on Robotics and Automation (ICRA), available online.
Interests
  • Artificial Intelligence
  • Reinforcement learning
  • Generative AI
  • Autonomous driving
  • Computer vision
  • Large Lanaguage Model
  • Game theory
Education

  • PhD in Computer Science and Engineering, 2023

    University of Connecticut

  • MS in Electrical and Computer Engineering, 2018

    Shanghai Jiao Tong University

  • BEng in Automation, 2015

    Nanjing University

Experience

 
 
 
 
 
Applied scientist intern
Amazon
May 2023 – Aug 2023 Sunnyvale, CA, USA
  • Crafted machine learning-driven solutions to efficiently handle time-series data characterized by sparse observations.
 
 
 
 
 
Research intern
Baidu USA
May 2020 – Dec 2020 Sunnyvale, CA, USA
  • Summarized exiting reinforcement learning methods and the state-of-art deep learning methods used in autonomous driving.
  • Built a prototype platform to train and test RL algorithms for autonomous vehicles in the Apollo platform and Amazon AWS.
 
 
 
 
 
Research assistant
University of Connecticut
Aug 2018 – May 2023 Storrs, CT, USA
  • Design a safe and scalable multi-agent reinforcement learning framework for the behavior planning and control of connected autonomous vehicles to improve traffic efficiency and safety.
  • Propose a stable and efficient reward reallocation algorithm to motivate cooperation for multi-agent reinforcement learning assuming all agents are self-interested.
  • Study the fundamental properties of the robust multi-agent reinforcement learning under adversarial state perturbations and propose a new objective and an algorithm to learn robust policy.
 
 
 
 
 
Research assistant
University of Michigan-Shanghai Jiao Tong University Joint Institute
Sep 2015 – Mar 2018 Shanghai, China
  • Proposed a flexible energy management approach to handle the uncertainties of weather and sizing in an isolated microgrid, which would not be influenced dramatically by different weather conditions.
  • Designed and fabricated high efficient bidirectional DC/DC converters to conduct and validate energy management approaches in a downsized system.

Research

*
All Autonomous Driving Other
Uncertainty Quantification on Collaborative Perception
Multi-agent collaborative perception has been proposed to leverage the viewpoints of other agents to improve the detection accuracy compared with the individual viewpoint. Recent research has shown the effectiveness of early, late, and intermediate fusion of collaborative perception, which respectively transmits raw data, output bounding boxes, and intermediate features, and the improved collaborative perception results will benefit the self-driving decisions of connected and autonomous vehicles (CAVs).
Uncertainty Quantification on Collaborative Perception
Game Theoretic Security Framework for Quantum Cryptography
We analyze quantum key distribution (QKD) protocols through a game theoretic framework. We make several contributions in this work. We propose a general-purpose framework allowing for a game-theoretic analysis of QKD protocols.
Multi-Agent Reinforcement Learning for Connected Autonomous Vehicles
Vehicle-to-vehicle (V2V) and Vehicle-to-Infrastructure (V2I) wireless connectivity is the next frontier in road transportation, which will greatly benefit the safety and reliability of autonomous cars. Information shared among autonomous vehicles provides opportunities to better coordination schemes and also raises novel challenges.
Multi-Agent Reinforcement Learning for Connected Autonomous Vehicles

Publications

Quickly discover relevant content by filtering publications.
Sanbao Su, Yiming Li, Sihong He, Songyang Han, Chen Feng, Caiwen Ding, Fei Miao (2023). Uncertainty Quantification of Collaborative Detection for Self-Driving. In ICRA 2023.

PDF Cite Project Video

Zhili Zhang, Songyang Han, Jiangwei Wang, Fei Miao (2023). Spatial-Temporal-Aware Safe Multi-Agent Reinforcement Learning of Connected Autonomous Vehicles in Challenging Scenarios. In ICRA 2023.

PDF Cite Project

Songyang Han, Sanbao Su, Sihong He, Shuo Han, Haizhao Yang, Fei Miao (2022). What is the Solution for State-Adversarial Multi-Agent Reinforcement Learning?. In arXiv.

PDF Cite

Songyang Han, Shanglin Zhou, Jiangwei Wang, Lynn Pepin, Caiwen Ding, Jie Fu, Fei Miao (2022). A Multi-Agent Reinforcement Learning Approach For Safe and Efficient Behavior Planning Of Connected Autonomous Vehicles. In arXiv.

PDF Cite Project Video

Songyang Han, He Wang, Sanbao Su, Yuanyuan Shi, Fei Miao (2022). Stable and Efficient Shapley Value-Based Reward Reallocation for Multi-Agent Reinforcement Learning of Autonomous Vehicles. In ICRA 2022.

PDF Cite Project

See all publications

Contact