A Novel Reinforcement Learning Architecture for Continuous State and Action Spaces

Table 1

Comparison of the best policies for the dribbling problem.

	SARSA A^2C	𝑄(𝜆)-learning
Algorithm type	Actor-Critic	𝑄(𝜆)-learning
Function approx.	RBFs	CMACs
States	Continuous	Continuous
Actions	Continuous	Discrete
Total learning time	10 minutes	24 hours 30 minutes
Average distance	25.45 meters	29.21 meters
Maximum distance	36.23 meters	39.0 meters