notesum.ai

Published at December 5

cs.RO

Released Date: December 5, 2024

Authors: Ruoyu Yao¹, Yubin Wang, Haichao Liu, Rui Yang, Zengqi Peng, Lei Zhu, Jun Ma

Aff.: ¹Hong Kong University of Science and Technology

Type	Method	NR-MR $\downarrow$	NR-CLS $\uparrow$	R-MR $\downarrow$	R-CLS $\uparrow$
Rule-based	PDM-Closed [18]	18.38%	67.72	13.24%	76.09
IL-based	UrbanDriverOL [39]	40.80%	51.67	44.49%	49.06
	RasterModel [7]	41.91%	50.65	40.44%	52.44
	PlanTF [10]	20.96%	72.56	32.72%	60.34
Diffusion-based	Diffusion-ES [51]	15.07%	77.54	12.87%	77.75
LMM-empowered (Ours)	CALMM-Drive ( $K$ =1)	14.34%	72.31	13.97%	72.87
	CALMM-Drive ( $K$ =3, fixed scorer)	15.81%	70.02	16.54%	71.09
	CALMM-Drive ( $K$ =3, complete)	13.24%	73.29	11.76%	73.71