About the Project
A core goal of Artificial Intelligence (AI) is the development of intelligent systems, or agents. Such agents could be virtual or physical (e.g., robots) and will need to be able to interact and coordinate with one another as well as humans. While the first such agents are starting to appear (e.g., assistants in smart phones, or robots in warehouses), further development of such intelligent agents and multi-agent systems (MASs) could revolutionize many aspects of society such as manufacturing, logistics, traffic systems and safety. A key challenge revolves around the task of decision making for such MASs; this is highly complex and requires smart approaches.
This project will aim to make both fundamental and applied contributions to this field of Multi-agent Sequential Decision Making (MSDM), which brings together MASs [9], reinforcement learning [7] and (decision-theoretic) planning [1]. In particular, we will investigate novel ways by which (deep) machine learning techniques [6, 4] can be used to improve the coordination of such MASs. These developed techniques will be evaluated in simulation in different application domains such as traffic light control [8] and robotics simulators [2]. We will also aim for a demonstration on team of real robots using a subset of the smARTlab’s 30 Turtlebots.
For this project, we are looking for highly motivated student with
• a background in artificial intelligence, math or physics, or a similar discipline;
• genuine, demonstrable, interest in decision making/reinforcement learning, as well as (deep) machine learning;
• excellent coding skills, and experience with python and C/C++; and
• good oral and writing skills.
In addition, familiarity with
• more advances formal frameworks such as partially observable MDPs [3] and decentralized POMDPs [5];
• planning methods such as Monte Carlo tree search;
• experience with contemporary machine learning frameworks such as Theano and Tensorflow are considered a plus.
For further information please contact: Dr. Frans Oliehoek [Email Address Removed]
References
[1] Craig Boutilier, Thomas Dean, and Steve Hanks. Decision-theoretic planning:
Structural assumptions and computational leverage. Journal of Artificial
Intelligence Research, 11:1–94, 1999.
[2] Daniel Claes, Philipp Robbel, Frans A. Oliehoek, Daniel Hennes, Karl Tuyls,
and Wiebe Van der Hoek. Effective approximations for multi-robot coordination
in spatially distributed tasks. In Proceedings of the Fourteenth International
Conference on Autonomous Agents and Multiagent Systems, pages
881–890, May 2015.
[3] Leslie Pack Kaelbling, Michael L. Littman, and Anthony R. Cassandra.
Planning and acting in partially observable stochastic domains. Artificial
Intelligence, 101(1-2):99–134, 1998.
[4] Yann LeCun, Yoshua Bengio, and Geoffrey Hinton. Deep learning. Nature,
521(7553):436–444, 2015.
[5] Frans A. Oliehoek and Christopher Amato. A Concise Introduction to Decentralized
POMDPs. Springer Briefs in Intelligent Systems. Springer, 2016.
[6] J. Schmidhuber. Deep learning in neural networks: An overview. Neural
Networks, 61:85–117, 2015.
[7] Richard S. Sutton and Andrew G. Barto. Reinforcement Learning: An Introduction.
The MIT Press, 1998.
[8] Elise Van der Pol and Frans A. Oliehoek. Coordinated deep reinforcement
learners for traffic light control. In NIPS’16 Workshop on Learning, Inference
and Control of Multi-Agent Systems, 2016.
[9] Gerhard Weiss, editor. Multiagent Systems. MIT Press, 2nd edition, 2013.
Continue with Facebook
