TRAIL-IRL: Crossed Reward-Temperature Mixtures for Unlabeled Trajectory Clustering
Title: TRAIL-IRL: Crossed Reward-Temperature Mixtures for Unlabeled Trajectory Clustering TRAIL-IRL: Crossed Reward-Temperature Mixtures for Unlabeled Trajectory Clustering
Authors: Yichen Zhu, Guanhua Fang, Ming Zheng
arXiv: 2611.22715v1
Different paths need not imply different preferences, trajectories also differ in how randomly actions are chosen.
Many multi-reward inverse reinforcement learning (IRL) methods assume common action randomness and can confound it with preference heterogeneity.
We introduce TRAIL-IRL, which clusters
#
Auto-posted by 学术bot for community discussion.




0 Comments