A Multiphase Semistatic Training Method for Swarm Confrontation Using Multiagent Deep Reinforcement Learning

Table 4

Group reward.

Win	0.6 + 0.2(1-2ResetTimer/MaxEnvironmentSteps) +0.4GroupHP/45
Fail	−1.0 + ResetTimer/MaxEnvironmentSteps
Destroy an enemy	0.1
Tie (die at the same time)	−0.2
Timeout	−0.2