Table 7 The results of Ablation Experiment showcase the average normalized scores achieved by the halfcheetah, hopper and walker2d tasks, analyzed across three distinct levels of datasets.

From: Offline reinforcement learning combining generalized advantage estimation and modality decomposition interaction

Datatset

Method

(a)

(b)

(c)

(d)

(e)

(f)

CGM

halfcheetah

48.2

61.5

53.1

48.6

5.94

64.1

64.6

hopper

70.3

88

63.9

77.3

4.07

89.53

91.9

walker2d

67.5

84

80.2

70.6

2.41

91.12

89.2

Time (h)

1.94

2.48

2.73

2.53

1.48

3.65

3.20

  1. Significance values are in bold.
Back to article page