RL in Social Robotics
Reinforcement Learning for Approaching Groups of People in a Socially Acceptable Manner.
With other postgraduates students, we used Reinforcement Learning to navigate a robot to a socially-acceptable point in a group of people while avoiding obstacles. My contribution was to adapt the problem of social-approach to be solved using RL with SARSA and Q-Learning. This project is on hold at the moment.
Example of one result using SARSA. The arrows indicate the movement the robot needs to do if it is at that discrete point, the red X are the the socially acceptable position for the robot to approach that group of people (inner circle).