pygmalionchen / drl_nav

Robot Formation Navigation in Gazebo with Deep Reinforcement Learning.

DRL_Nav

Robot Formation Navigation in Gazebo with Deep Reinforcement Learning.

---

DDPG Formation Navigation：

The DDPG_formation_test.py and DDPG_formation_train.py are files for multi robots navigation. However, they still be idiot.

• The robots4_formation.launch and robots8_formation.launch are for my previous tests.
• The multi_goal.launch belongs to the predecessor. The corresponding training and testing scripts are saved on the disk.

---

System Setup:

Add New Env:

Register the class in the tf_rl/__init__.py and import the env script in the tf_rl/envs/__init__.py.

World Files Introduction:

The main world files used in our project are described below. If you want to modify the world file, you should

• ddpg3Block.world: 3 cinder blocks lying on the points (-2,-2) (0,0) (2,2). The rectangle wall from (-6,-9) to (8.5,6).
• maddpg1block.world: 1 jersey_barrier lying on the middle. Approximatly from (-0.3,0) to (3,-4). The rectangle wall from (-6,-9) to (8.5,6).
• maddpg111.world: 3 jersey_barrier lying in parallel. From (-6,-4) to (2.2,-7), (0,0) to (8.5,-1.3), (-6,2) to (2.1,3). The rectangle wall from (-6,-9) to (8.5,6).
• maddpgFree.world： Free environment. The rectangle wall from (-6,-9) to (8.5,6).
• maddpgblock.world and wallblock.world need to be evaluated later.