Conditional neural models learn goals early in action imitation
Conditional Neural Processes for Self-Supervised Goal Discovery in Action Imitation
RoboticsArtificial Intelligence
Summary
Learning to imitate actions by understanding their goals is a key skill for robots. This paper shows that adding an attention mechanism to a neural model helps it quickly recognize the goal behind observed actions without needing labeled data. The model builds strong goal representations from just part of an action sequence, much like how infants learn by watching. This approach could help robots learn tasks on their own by watching others.
What this means in practice
- •For robot developers: Develop robots that autonomously infer task goals by observing limited action sequences without requiring manual annotations.
- •For game ai designers: Create agents that learn player goals from partial demonstrations to improve interactive and adaptive gameplay experiences.
Authors
Marco Gabriele Fedozzi, Yukie Nagai, Francesco Rea, Alessandra Sciutti
Abstract
Inspired by infants' ability to learn novel tasks by imitating others, in this work we evaluate to what extent Conditional Neural Processes can autonomously learn representations of actions that are informative about their underlying goal. We show that the addition of an attention mechanism in the hidden space fosters the emergence of robust goal representations early on in the observation of an action sequence, making it a promising model for a robotic self-supervised general learner.