Single ML agent meets Dummy agent

The yellow agent learns a policy, while the cyan agents applies the Dummy behavior instead, which just makes it move at full speed forward.

The policy computes (linear) accelerations. Because Dummy will always move at the same constant speed, the policy does not need to know the its speed.

🟧 -> 🔶

• Continous actions
  • SAC
  • PPO
  • PPO trained on starting from opposing sides
• Discrete actions
  • Multi-binary action space
  • Discrete action space