notesum.ai

Published at October 21

Diverse Policies Recovering via Pointwise Mutual Information Weighted Imitation Learning

cs.CV

cs.AI

Released Date: October 21, 2024

Authors: Hanlin Yang¹, Jian Yao², Weiming Liu², Qing Wang², Hanmin Qin², Hansheng Kong², Kirk Tang², Jiechao Xiong², Chao Yu¹, Kai Li³, Junliang Xing⁴, Hongwu Chen², Juchao Zhuo², Qiang Fu², Yang Wei², Haobo Fu²

Aff.: ¹Sun Yat-sen University, Guangzhou, China; ²Tencent AI Lab, Shenzhen, China; ³Institute of Automation, Chinese Academy of Sciences, Beijing, China; ⁴Tsinghua University, Beijing, China

Arxiv: https://arxiv.org/abs/2410.15910v1

Style	Metrics	BC	CBC	BC-PMI
Class 1 (Red)	DTW	88.590 $\pm$ 9.329	8.130 $\pm$ 0.732	7.511 $\pm$ 0.545
	ED	133.100 $\pm$ 20.210	22.039 $\pm$ 1.383	21.591 $\pm$ 1.091
	KL	8.037 $\pm$ 0.870	0.044 $\pm$ 0.006	0.037 $\pm$ 0.004
Class 2 (Green)	DTW	45.729 $\pm$ 5.293	7.695 $\pm$ 0.814	7.631 $\pm$ 0.821
	ED	88.180 $\pm$ 7.541	35.355 $\pm$ 2.691	35.777 $\pm$ 2.912
	KL	1.111 $\pm$ 0.072	0.043 $\pm$ 0.003	0.037 $\pm$ 0.003
Class 3 (Orange)	DTW	77.619 $\pm$ 7.812	18.972 $\pm$ 2.323	11.576 $\pm$ 1.788
	ED	130.839 $\pm$ 10.322	107.833 $\pm$ 8.213	74.267 $\pm$ 8.002
	KL	16.839 $\pm$ 1.022	0.271 $\pm$ 0.031	0.135 $\pm$ 0.014
Class 4 (Blue)	DTW	69.527 $\pm$ 10.924	7.755 $\pm$ 1.832	7.577 $\pm$ 1.051
	ED	97.066 $\pm$ 14.239	27.230 $\pm$ 2.110	26.749 $\pm$ 2.347
	KL	36.238 $\pm$ 2.981	0.401 $\pm$ 0.061	0.219 $\pm$ 0.019