Open WangJuan6 opened 2 years ago
I am not completely sure I understand your issue. The robot's action is being physically simulated, so if the robot uses a forward action at timestep 0 and a turn-left action at timestep 1, it will still inherit some forward momentum at timestep 1 (like in the real world). Is this the position offset you are referring to?
@mjlbach Would you mind sharing your approach or maybe code snippet to build a global point cloud?
Hi @ChengshuLi :
Thanks for your reply.
For the first graph, I use the husky robot and set the robot using forward action at every timestep, the result seems correct.
I just found if I use the turtlebot robot and set the robot using forward action at every timestep, the result is not correct:
The husky robot always collision with other objects and change the position of other objects. I can not initialize it to a good position by env.simulator.robots[0].set_position([10,10,0]). So, I use turtlebot robot in the second and third graphs, and setting the robot uses turn-left action at every timestep. The second graph is the result using the rotation matrix and transformation matrix. The third graph is the result of using the rotation matrix only. So, I guess there is some position offset of turtlebot robot. Thanks, Juan
Hi @ChengshuLi: I have other questions about the environment, can you answer me?
Thanks, Juan
Dear iGibson team: I'm trying to build the global point cloud graph. I use the position and orientation to calculate the rotation matrix and transformation matrix.
When the agent performs the forward action, the result is as follows:
When the agent performs the rotation action, the result is as follows:
Obviously, the second graph has offset.
I try to use rotation matrix only when the agent performs rotation action, the result is as follows:
I guess the error is caused by the position offset of the rotation action. So, how can I eliminate this position offset?
Thanks,
Juan