Experimentation & Main Training Runs

[KevinBian107]

We need to run the following:

Single Agent v.s. Bots

• PPO/Smart
• PPO/Aggresive
• PPO/Defensive
• PPO/Dodge
• SAC/Smart
• SAC/Aggresive
• SAC/Defensive
• SAC/Dodge

Cycle learning

Run with a pre-set (best) eseuqneces of bot to fight against). If it work, we will include more of this section, if not, we will include one result for reference.

• Compare with single PPO baeline fighting against bots

Single Agent v.s. Agent

We can transfer learning this from cycle learning or bots learning.

• PPO/PPO
• SAC/SAC
• PPO/SAC

Team Player Mode

• Agent v.s. bots different combo
• Agent v.s. agent different combo
• PPO 2A/Smart 2B
• PPO 2A/Combo of 3B (should include at least one defensive)
• PPO 2A/PPO 2A

Special Notice

1. Battlefield with wall should only use smart bots (iff)
2. Set a hyperparameter
3. Set rewards