Distributed policy

In these following notebooks, we train a distributed policy shared between the two agents.

As for the previous notebooks, the policy computes (linear) accelerations. We test different observations spaces, action spaces and training algorithms.

🟧 -> 🔶

• Parallel SAC
  • Perceiving neighbor position and speed
  • Without perceiving neighbors
  • Perceiving only neighbor position
  • Perceiving only neighbor speed
• Parallel PPO with discrete actions
  • Training
  • Evaluation
• BenchMARL
  • Continuos actions with MASAC
  • MASAC with global state critic
  • MAPPO
  • MAPPO starting from opposite sides
  • MASAC with discrete actions