3 ms·
They don't seem to be using end-to-end deep reinforcement learning. I think the robot movements will become smooth and organic once they replace the current spe
by visarga 11y ago
They don't seem to be using end-to-end deep reinforcement learning. I think the robot movements will become smooth and organic once they replace the current specialized approach with more general algorithms.
Robotics is a core application of reinforcement learning as it deals with an embedded agent in a feedback look with the exterior. That's why DeepBrain chose games to test their algorithm, because in a game the player is embedded in an artificial world which is much easier to use in the trial and error learning phase. Logically, after mastering Atari games, at some point they should focus on robots that function in the real world. It helps that Deep Brain and Boston Dynamics are owned by the same company.