Multirobot Payload Transport with Deep Reinforcement Learning

In this project several centralized and decentralized methods for Multirobot Payload transport are explored there are two scenarios for the robots, A closed space with a target and a narrow passage way for the robots to carry payload to. The algorithms experimented are

The code can be found in the "examples" directory, Go ahead and explore all the methods you can find here!

• Cenralized:

  • Proximal Policy Optimization(PPO)
  • Deep Determinstic Policy Gradients(DDPG)
  • Evolution Guided Policy GradientsPPO-ERL

• Distributed :

  • MAPPO
  • MADDPG
  • Neurocoevolution
    • Hall Of Fame

Demo:

• PPO-ERL:

• PPO:

• Evolutionary methods NES:

Project simulation is in PyBullet environment and depends on Generic URDF robots. I'm thankful to Max Spahn for his prompt responses and help in building the simulation environment.