Hello, we said that we can use policy parameters to enable DMP to generate arbitrary shaped tracks. For example, when an object is grabbed from point A and moved to point B, the corresponding track is the Demonstration track.
Now, if you want to move point A object to point C, should you use the noise exploration you mentioned here to train the Demonstration track, or directly generate a new trajectory with DMP, and directly generalize to point C?
What is the difference between the two methods?
Thank you.
Hello, we said that we can use policy parameters to enable DMP to generate arbitrary shaped tracks. For example, when an object is grabbed from point A and moved to point B, the corresponding track is the Demonstration track. Now, if you want to move point A object to point C, should you use the noise exploration you mentioned here to train the Demonstration track, or directly generate a new trajectory with DMP, and directly generalize to point C? What is the difference between the two methods? Thank you.