Browse Teach Curate

Deep Reinforcement Learning for Goal Recognition with Actor-Critic Policy Networks

An approach to inferring an agent's goal from observed actions by learning one actor-critic policy network per candidate goal from unstructured data and scoring hypotheses over continuous policy representations. A student learns how deep reinforcement learning replaces hand-built symbolic domain theories in goal recognition and extends it to continuous domains.

B9 Text

Goal Recognition using Actor-Critic Optimization Ben Nageris1, Felipe Meneguzzi2, Reuth Mirsky1, 3

DRACO (Deep Recognition using Actor-Critic Optimization) is a deep-reinforcement-learning method for goal recognition - inferring which goal an observed agent is pursuing from a sequence of its actio…