Compositional Reinforcement Learning
In this line of work, we are interested in techniques that allow an agent to leverage past knowledge to solve new tasks quickly. In particular, we focus on how agents can acquire behaviours that can be combined to generate interesting, novel abilities. One particular focus is on applying Boolean operators to learned behaviours to generate provably optimal solutions to new tasks. Not only are these approaches human-understandable, but they result in a combinatorial explosion in an agent's abilities, which is key to tackling the multitask or lifelong learning setting.
|
G. Nangue Tasse, S. James, B. Rosman. A Boolean Task Algebra for Reinforcement Learning. Advances in Neural Information Processing Systems (NeurIPS), December 2020. |
G. Nangue Tasse, S. James, B. Rosman. Logical Composition for Lifelong Reinforcement Learning. 4th Lifelong Learning Workshop at ICML, July 2020. |
B. van Niekerk, S. James, A. Earle, B. Rosman. Composing Value Functions in Reinforcement Learning. International Conference on Machine Learning, June 2019. |
A. Earle, A. Saxe, B. Rosman. Hierarchical Subtask Discovery with Non-Negative Matrix Factorization. Proceedings of the Sixth International Conference on Learning Representations, April 2018. |
A. Saxe, A. Earle, B. Rosman. Hierarchy Through Composition with Multitask LMDPs. International Conference on Machine Learning, August 2017. |
Skills & Symbols
Here we focus on learning abstract representations, which we believe are an important component if we are ever to apply reinforcement learning to the real world. In particular, we focus on skill- and symbol-discovery, as well as the interplay between the two. We have applied our approaches to challenging pixel-based tasks that require high-level planning, and have shown that symbolic representations can be learned directly from low-level sensor data.
|
S. James, B. Rosman, G. Konidaris. Learning Portable Representations for High-Level Planning. International Conference on Machine Learning, July 2020. |
S. James, B. Rosman, G. Konidaris. Learning Object-Centric Representations for High-Level Planning in Minecraft. Object-Oriented Learning (OOL): Perception, Representation, and Reasoning. Workshop at ICML, July 2020. |
O. Marom, B. Rosman. Zero-Shot Transfer with Deictic Object-Oriented Representation in Reinforcement Learning. Advances in Neural Information Processing Systems (NeurIPS), December 2018. |
T. Taniguchi, E. Ugur, M. Hoffmann, L. Jamone, T. Nagai, B. Rosman, T. Matsuka, N. Iwahashi, E. Oztop, J. Piater, F. Worgotter. Symbol Emergence in Cognitive Developmental Systems: a Survey. IEEE transactions on Cognitive and Developmental Systems, January 2018. |
P. Ranchod, B. Rosman, G. Konidaris. Nonparametric Bayesian Reward Segmentation for Skill Discovery Using Inverse Reinforcement Learning. IEEE/RSJ International Conference on Intelligent Robots and Systems, September 2015. |
Theory of Mind
Our research here focuses on modelling external agents in an environment. These agents may be other robots or humans that have their own goals or intentions that are not directly observable by our agent. Inferring this information through observation or communication can allow agents to better collaborate to achieve the required task optimally. In particular, as robots become more ubiquitous in the real world, human-robot interaction will be key to ensuring productive and safe environments.
|
O.C. Görür, B. Rosman, S. Albayrak.. Anticipatory Bayesian Policy Selection for Online Adaptation of Collaborative Robots to Unknown Human Types. International Conference on Autonomous Agents and Multiagent Systems, May 2019. |
O.C. Görür, B. Rosman, F. Sivrikaya, S. Albayrak. Social Cobots: Anticipatory Decision-Making for Collaborative Robots Incorporating Unexpected Human Behaviors. ACM/IEEE International Conference on Human-Robot Interaction, March 2018. |
O.C. Görür, B. Rosman, G. Hoffman, S. Albayrak. Toward Integrating Theory of Mind into Adaptive Decision-Making of Social Robots to Understand Human Intention. Workshop on the Role of Intentions in Human-Robot Interaction at the International Conference on Human-Robot Interaction, March 2017. |
P. Hernandez-Leal, M. Taylor, B. Rosman, E. L. Sucar, E. Munoz de Cote. A Bayesian approach for Learning and Tracking Switching, Non-stationary Opponents. Autonomous Agents and Multiagent Systems, May 2016. |
P. Hernandez-Leal, M. Taylor, B. Rosman, E. L. Sucar, E. Munoz de Cote. Identifying and Tracking Switching, Non-stationary Opponents: a Bayesian Approach. Workshop on Multiagent Interaction without Prior Coordination (MIPC), at AAAI, February 2016. |
B. Rosman, M. Hawasly, S. Ramamoorthy. Bayesian Policy Reuse. Machine Learning Journal, 104(1), pp. 99-127, June 2016. |
RoboCup
Our group competes in the RoboCupSoccer 3D Simulation, in which a team of 11 simulated Nau robots compete in a game of football against other teams from around the world. Our focus here is both on improving the low-level control of individual robots and incorporating high-level, multi-agent decision making into the team's strategy.