“Reinforcement Learning in Traffic Signal Control”
In this talk, I will present recent results from the TensorCell research group on reinforcement learning for traffic signal control. Independent agents controlling traffic at different intersections are trained using the IDQN and IPPO algorithms to make the traffic flow as efficient as possible. Experiments on realistic Melaka traffic scenarios simulated in the SUMO traffic simulator and using the RESCO framework yielded very good results, outperforming not only centralized control by single agents trained with DQN and PPO algorithms, but also the Webster algorithm, which is widely used in practical traffic engineering applications.
Bio
Dr Paweł Gora – Scientist, IT specialist, philantropist and entrepreneur working mainly on the applications of AI (especially in transportation and medicine) and quantum computing. Graduated from the Faculty of Mathematics, Informatics and Mechanics of the University of Warsaw (M.Sc. in Mathematics, M.Sc. in Computer Science, and Ph.D. in Computer Science).
He is the founder and CEO of the “Quantum AI Foundation” http://www.qaif.org, which aims to support education and research in new technologies (especially AI and quantum computing) by organizing meetups (e.g., Warsaw.AI and Warsaw Quantum Computing Group), workshops on quantum computing, conferences, competitions and hackathons, as well as publishing newsletters (Warsaw.AI News and QPoland Newsletter) and moderating social media groups and profiles related to AI and quantum computing and ITS. He is also the Chairperson of the Board of QWorld and Coordinator of one of its local QCousins, QPoland. He also collaborates with several startups serving as a technical and business advisor.
He has more than 15 years of experience in the field of Intelligent Transportation Systems (ITS). He has worked on projects related to modeling and simulation of transport networks, traffic management systems, carpooling/vanpooling services, bike-sharing services, electric vehicles, connected and autonomous vehicles, intelligent parking systems, optimization of logistics. He built a microscopic traffic simulation software, Traffic Simulation Framework, and used it for experiments with AI (graph neural networks, reinforcement learning) to optimize traffic signal settings in the TensorCell research group, which he founded in 2017 and which aims to optimize complex processes using AI.
He has received several awards, e.g., “Lider ITS” award for the best R&D work in the ITS domain in Poland. “MIT Technology Review” recognized him as one of 10 Top Polish Talents in the “MIT Innovators Under 35” competition. In 2017, he was also placed on the list “NEW EUROPE 100” of emerging technology stars in Eastern Europe.
In the past, he worked as a software engineering intern or research intern at Microsoft, Google, CERN and IBM Research, was also a representative of Poland in the Management Committee of the COST Action “Wider Impacts and Scenario Evaluation of Autonomous and Connected Transport”, and a member of the Council for Digitalization.