I switch Panda Goal Reaching example to ENU that is the new default coordinate system on Webots R2022a.
Since we only need to consider the relative position, we don't need to modify our code except the wbt file.
Check if it works correctly without training:
Download the pretrained agent into the following folder: /deepworlds/examples/panda/panda_goal_reaching/controllers/robot_supervisor_manager/tmp/ddpg.
Please add message = [message[0], message[2], -message[1]] between these two line.
The pretrained agent is trained in NUE, so we need a coordinate transformation (ENU(x, y, z) -> NUE(x, z, -y)).
You should be able to see that the agent works correctly.
I switch Panda Goal Reaching example to ENU that is the new default coordinate system on Webots R2022a. Since we only need to consider the relative position, we don't need to modify our code except the wbt file.
Check if it works correctly without training:
/deepworlds/examples/panda/panda_goal_reaching/controllers/robot_supervisor_manager/tmp/ddpg.Directories structure:
Please add
message = [message[0], message[2], -message[1]]between these two line. The pretrained agent is trained in NUE, so we need a coordinate transformation (ENU(x, y, z) -> NUE(x, z, -y)).You should be able to see that the agent works correctly.