RL_mobile_manip

I'm still Testing... ㅠㅠ

Please wait for month...!

Reinfore Learning / Mobile manipulator Task Planning implementaion by using DDPG

Experiment

'ros-noetic', 'ubuntu 20.04'

- Simulation Scence (Coppeliasim)

- IRM structure about sampled Grasp Pose from Point cloud (Red is High value)

- RL sequence visulazation using smach

Excute demo program

1. excute 'roscore' for CoppeliaSim rosInterface.
2. turn on 'Panda_Husky_test.ttt' for test.
3. run 'roslaunch TAMP_interface interface.launch' on your terminor.
4. run 'rosrun smach_viewer smach_viewer.py' you can see whole process on smach viewer!!

topic name

• PointCloud2 : '/k4a/depth_registered/points'
• MakerArray : 'ws_IRM'

TODO

1. make IRM with numerical method is inefficient --> get Distribuiton using GAN (fast sampling)
2. How to use Quality Measure??
3. study RL / ML for efficient implementation.

Reference

• Vahrenkamp, N., Asfour, T., Metta, G., Sandini, G., & Dillmann, R. (2012). Manipulability analysis. 2012 12th IEEE-RAS International Conference on Humanoid Robots (Humanoids 2012)

• https://github.com/jkw0701/robocare_motion_planner.git

• https://github.com/beomjoonkim/adversarial_actor_critic.git

• https://gitlab.com/Simox/simox.git

..

if you find good reference fir project ... pleaze share me... [email protected]