MORE | Summer 2021

Perspective Variant Preference-Based Learning with Quadcopter Swarms

Preference-based learning allows for non-expert users to interact with reinforcement learning agents in a way that they can shape their behaviors. This work intends to push the current implementations of preference-based learning forward by introducing binocular vision, multiple perspectives of the agent performing the actions, and multi-agent environments. These three improvements on the existing implementations will allow for preference-based learning to be more robust and useful in portraying an accurate representation of the actions that the user wishes for the agent to perform.