Event

Proposes a Meta-Learning Framework for HRI Pattern Recognition and Adaptive Control

1 reports · 1 sources

Nuo Chen and Ya-Jun Pan propose a meta-learning-based human-robot interaction framework, using a meta-learning-trained neural network to recognize two modes—active guidance and passive following—without manually setting switching thresholds; the paper claims that after introducing force-related parameters into the controller, passive mode recognition rate rose from 90.1% to 99.1%, and active mode from 79.8% to 97.1%. The accompanying adaptive meta-learning controller is used to compensate for human disturbances and adapt online to different interaction forces; a multi-trajectory meta-learning algorithm jointly learns environment base representations and force-related parameters to handle unmodeled dynamics; the authors claim experiments verified its robust control and reliable pattern recognition performance.