August 21st, 2021
Recent Advances in Reinforcement Learning for Human-AI Collaboration
Recent advances in reinforcement learning (RL) have led to remarkable performance of AI systems in challenging application domains, e.g., robotics and game play (Go, etc.). However, these feats have largely been limited to well-specified tasks with known dynamics and predictable outcomes. These limitations can be addressed by designing AI systems that emerge from the complementary abilities of humans and machines by enabling close collaborations between them. For instance, in autonomous driving, an AI auto-pilot could hand over control to the human driver in safety-critical situations. To enable such collaboration, there has been a surge of interest in developing novel RL techniques that effectively and efficiently learn with-and-from people in complex real-world environments.
In this tutorial we provide an overview of the recent advances in designing RL techniques for human-AI collaborations. The contents are loosely based on our workshop "Human-AI Collaboration in Sequential Decision-Making" at ICML'21.
Our tutorial consists of four parts:
[Sources:David Silver's RL lectures]
Optimizing teamwork via adaptation
[Sources: A. Ghosh et al. AAMAS 2020 paper / J. Förster et al. AAMAS 2018 paper]
Ensuring safety via shielding and shared autonomy
[Sources: M. Alshiekh et al. AAAI 2018 paper /
J. Inala et al. Human-AI Collaboration WS 2021 paper /
H. M. Sajjad Hossain et al. Human-AI Collaboration WS 2021 paper]
Improving collaboration via action advising and teaching
[Sources: L. Torrey and M. Taylor AAMAS 2013 paper / O. Amir et al. IJCAI 2016 paper /
S. Omidshafiei et al. AAAI 2019 paper]
Max Planck Institute for Software System
Unversity of Vienna
If you have any questions or comments please don't hesitate to contact us at email@example.com.