<ul><li>Imitation learning often assumes demonstrations are close to optimal based on an unknown cost function.</li><li>Focused satisficing approach seeks to surpass the demonstrator's aspiration levels without explicitly learning them.</li><li>Using margin-based objective in deep reinforcement learning, it improves imitation learning by imitating highest quality demonstrations.</li><li>Experimental results show improved policy focus and competitive true returns on various environments.</li></ul>

Imitation Learning via Focused Satisficing

Discover more