Avoiding Chatter in an Online Co-Learning Algorithm Predicting Human Intention

摘要

Chatter can happen when an online learning algorithm is used by a robot to predict human intention while interacting with a human subject. When chatter happens, the learning algorithm continually changes its prediction, without reaching a constant prediction of human intention. Using the Rescorla-Wagner model for human learning, we analyze an expert based online learning algorithm and identify an invariant set in the state and parameter space where chatter will occur. Based on the chatter analysis, we also propose an improved expert based learning algorithm where the invariant set does not exist so that chatter can be avoided.