End-to-End Autonomous Exploration with Deep Reinforcement Learning and Intrinsic Motivation

Table 7

Experiment results of “noisy-TV.”

Environment	Method	Reward	MER (%)	IQRE
Maze-1	ICM + scratch	315.62	53.86	N/A
	Ours + scratch	582.74	100	7.58
	ICM + fine-tuning	374.52	64.05	N/A
	Ours + fine-tuning	586.43	100	8.67
Maze-2	ICM + scratch	279.68	48.71	N/A
	Ours + scratch	565.32	100	6.93
	ICM + fine-tuning	317.54	56.18	N/A
	Ours + fine-tuning	566.73	100	7.75
Maze-3	ICM + scratch	362.49	63.28	N/A
	Ours + scratch	577.86	100	7.69
	ICM + fine-tuning	305.47	54.72	N/A
	Ours + fine-tuning	572.63	100	8.12