I was wondering whether it is possible to use a feature or signal for reward calculation without explicitly having the trajectory optimizer using it as part of the state feedback.
For example, if I want to use an object's pose to calculate a reward, but only want to use the robot's joints for the feedback controller, how would I go about doing that? If I do not include the object's pose as a state, the algorithm throws an error if I want to use it for reward computation.
Hello,
I was wondering whether it is possible to use a feature or signal for reward calculation without explicitly having the trajectory optimizer using it as part of the state feedback.
For example, if I want to use an object's pose to calculate a reward, but only want to use the robot's joints for the feedback controller, how would I go about doing that? If I do not include the object's pose as a state, the algorithm throws an error if I want to use it for reward computation.