notesum.ai

Published at December 10

Contractive Dynamical Imitation Policies for Efficient Out-of-Sample Recovery

cs.LG

cs.RO

stat.ML

Released Date: December 10, 2024

Authors: Amin Abyaneh¹, Mahrokh G. Boroujeni², Hsiu-Chin Lin, Giancarlo Ferrari-Trecate

Aff.: ¹McGill University; ²Ecole Polytechnique Fédérale de Lausanne (EPFL)

Arxiv: http://arxiv.org/pdf/2412.07544v1

Refer to caption

Expert	LASA-2D		LASA-4D		Robomimic-6D		Robomimic-14D
Metric	MSE	soft-DTW	MSE	soft-DTW	MSE	soft-DTW	MSE	soft-DTW
SNDS	0.02 $\pm$ 0.01	0.72 $\pm$ 0.14	0.03 $\pm$ 0.01	1.04 $\pm$ 0.19	0.65 $\pm$ 0.55	1.26 $\pm$ 0.70	2.42 $\pm$ 1.37	4.15 $\pm$ 0.92
BC	0.04 $\pm$ 0.02	0.98 $\pm$ 0.12	0.05 $\pm$ 0.03	1.48 $\pm$ 0.16	0.56 $\pm$ 0.32	1.88 $\pm$ 0.20	1.75 $\pm$ 0.22	4.86 $\pm$ 0.58
SDS-EF	0.03 $\pm$ 0.01	0.85 $\pm$ 0.13	0.05 $\pm$ 0.02	1.10 $\pm$ 0.15	0.50 $\pm$ 0.28	1.01 $\pm$ 0.66	3.30 $\pm$ 0.75	5.65 $\pm$ 0.58
SCDS	0.02 $\pm$ 0.01	0.65 $\pm$ 0.05	0.03 $\pm$ 0.01	0.72 $\pm$ 0.12	0.56 $\pm$ 0.22	1.05 $\pm$ 0.37	1.68 $\pm$ 0.45	4.10 $\pm$ 0.40
SNDS	2.73 $\pm$ 1.67	6.91 $\pm$ 1.46	3.65 $\pm$ 2.12	9.85 $\pm$ 0.63	1.58 $\pm$ 0.93	3.27 $\pm$ 1.88	6.88 $\pm$ 2.16	12.17 $\pm$ 2.70
BC	8.63 $\pm$ 4.05	16.25 $\pm$ 5.27	19.25 $\pm$ 7.34	27.48 $\pm$ 6.83	11.05 $\pm$ 7.41	19.94 $\pm$ 9.57	44.98 $\pm$ 15.11	37.82 $\pm$ 14.13
SDS-EF	1.78 $\pm$ 0.34	7.13 $\pm$ 1.51	2.33 $\pm$ 0.47	10.21 $\pm$ 1.98	1.15 $\pm$ 0.81	2.67 $\pm$ 1.40	11.10 $\pm$ 2.10	10.44 $\pm$ 1.66
SCDS	0.32 $\pm$ 0.15	1.72 $\pm$ 0.54	1.09 $\pm$ 0.21	2.58 $\pm$ 0.30	0.89 $\pm$ 0.26	1.71 $\pm$ 0.53	2.68 $\pm$ 0.65	6.27 $\pm$ 0.48
$\mathcal{L}_{\text{ub}}^{\text{MSE}}$	0.49 $\pm$ 0.07		1.33 $\pm$ 0.14		1.43 $\pm$ 0.22		2.91 $\pm$ 0.45