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

Table 3

Individual reward.

Win (alive)	0.6 + 0.6(HP/FullHP) −0.2ResetTimer/MaxEnvironmentSteps
Win (deceased)	0.3
Dead	−0.3
Destroy an enemy	0.15
Collision with tank	−0.05
Collision with wall	−0.5