Hierarchical Reinforcement Learning For in Hand Manipulation
Hierarchical Reinforcement Learning For in Hand Manipulation
Embedded Files
I am exploring how HRL can be used to speed up learning in in hand manipulation. Specifically I'm looking at how much we should break down the task of in hand manipulation to improve training time and performance in simulation and the real world
I am exploring how HRL can be used to speed up learning in in hand manipulation. Specifically I'm looking at how much we should break down the task of in hand manipulation to improve training time and performance in simulation and the real world
Currently I have trained policies capable of translating and rotation objects, as well as establishing contact at useful points. I am working on combing these into more complex behaviors for upcoming work
Currently I have trained policies capable of translating and rotation objects, as well as establishing contact at useful points. I am working on combing these into more complex behaviors for upcoming work
Shown here is a rotation trial using one of our trained policies. The agent is able to learn to rotate the objects in place using from any starting position within the reachable space of the fingers. Combining this with our sliding controller from the previous work will allow us to accomplish more complex tasks.
Page updated
Google Sites
Report abuse