Object tracking video using Reinforcement learning

How to launch original env

First launch the simulator

  roslaunch ur_robotiq_gazebo gym.launch
  

And run the training launch

  roslaunch ur_training default.launch
  

Conveyer GAZEBO env

First launch the gazebo and gym interface and node publishing block point.

 roslaunch ur_robotiq_gazebo conveyer_gym.launch --screen
 

Run the RL algorithms and unpause the GAZEBO

  roslaunch ur_training default.launch
  

Latest block’s point:

rostopic echo /target_blocks_pose

Total block’s points:

rostopic echo /blocks_poses 

How to launch REINFORCE algorithm

First launch the simulator

roslaunch ur_robotiq_gazebo conveyer_gym.launch controller:=vel --screen gui:=false
  

And load the parameters and launch python file for reset

roslaunch ur_reaching reinforcement.launch
  

And start the learning algorithm

python reinforcement_main.py 
  

And unpause physics of GAZEBO simulator

 rosservice call /gazebo/unpause_physics "{}"
 

How to launch PPO+GAE algorithm

First launch the simulator including loading the parameters and GAZEBO Excution func

roslaunch ur_robotiq_gazebo conveyer_gym.launch --screen gui:=false

And start the learning algorithm

  python ppo_gae_main.py
 

How to use the RLkit

RLkit is reinforcement learning framework based on rllab

Run GAZEBO simulator and gazebo_execution

First launch the simulator including loading the parameters and GAZEBO Excution func

roslaunch ur_robotiq_gazebo conveyer_gym.launch --screen gui:=false

Training

And start the SAC learning algorithm based on RLkit

  python rlkit_sac_main.py
 

And unpause physics of GAZEBO simulator

 rosservice call /gazebo/unpause_physics "{}"
 

After training, you may find the pickled files on the rlkit/data folder.

you can easily see the results through selecting the generated folder about training like follwing:

File truncated at 100 lines see the full file