Imitation Learning and Response Facilitation in Embodied Agents

摘要: Imitation is supposedly a fundamental mechanism for humans to learn new actions and gain knowledge about another’s intentions. The basis of this behavior seems be direct influencing the motor system by perceptual system, affording fast, selective enhancement response already in repertoire (response facilitation) as well learning delayed reproduction (true imitation). In paper, we present an approach attain these capabilities virtual embodied agents. Building upon computational control model, our connects visual representations observed hand arm movements graph-based commands. Forward inverse models are employed allow both fast mimicking responses imitation learning.