Learning Acrobatic Flight from Preferences

Colin Merk, Ismail Geles, Jiaxu Xing +3

Preference-based reinforcement learning (PbRL) enables agents to learn control policies without requiring manually designed reward functions, making it well-suited for tasks where…