Transfer Reinforcement Learning across Environment Dynamics with Multiple Advisors
This work belongs in the realm of Transfer Learning in RL, which generally aims at allowing an RL agent to learn a new task faster by exploiting already acquired knowledge of previous tasks. The presented method proposes to exploit not only one past experience in a particular previous task (one advisor), but several experiences in several slightly different previous tasks (multiple advisors). The agent can then follow the advisors’ advice when the majority agrees on what to do, and deflect from it when the advisors disagree.
I’m a PhD student at the Artificial Intelligence lab of the VUB. The main focus of my research is Reinforcement Learning, specifically how RL agents learn from human input. Since January 2019, I’m working on a 4 years project aiming at allowing a motorized wheelchair to learn to navigate in an indoor environment while receiving advice and feedback from a user (sitting in the wheelchair).