Closed Gloriabhsfer closed 4 years ago
What control mode are you using? Are you using the solve_ik
function or the get_path
function to get to the target?
Also, I'm not sure what you are doing in line 3 here:
robot_done=self.path.step()
self.pr.step()
self.path.set_to_start() # It's a little unusual to have this line here (in the loop)
Best, Stephen
Hi Stephen:
Sorry for the late update. My control mode is inverse kinematic. I use get_path
inside the controller. My goal is to let the end-effector learning how to find a path to a random point inside the whole workspace.
I use bullet 2.78 engine.
Because I use DQN and it is discrete action space algorithm.My strategy is make the action space become 6 range for -x,+x,-y,+y,-z,+z and everystep I will get a action base the policy DQN gave me and let my end-effector move to this position base on IK. I firstly try the solve_ik
but when I read the example for end-effectro control, it seemsget_path
can use some algorithm like SBL,RRT( which is inside the arm.py code) So I choose to use this function as my low level controller.
The line you quote is I need to decided the whether the robot is go to the target position or not. If it is done, I can calculate the reward function for this step and decided if I need more action to get the next step.
Best,
Gloria
This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.
Hi Stephen: Thanks for the awsome package. I am using the package to train a reinforcement learning model with UR5 in Vrep. But when I update my go position for reinforcement learning ( I use DQN in this case), it cannot let me go to the target position for each step. My plan is to update my end effector through each step for one episode. But it will break the simulation and return my some error like this:
If the controller cannot find a path, I make the episode end and let it go to another attemption. I am not very sure about your backend structure, could you tell me where I need to adjust in my low-level control so I can realize a single follow a point end-effector controller?
Best, Yu Zhang