The goal of this post is to capture my progress in learning of object reaching on my iCub.
Suppose iCub stands in the front of the object, which is placed on a table, so the robot can reach it without moving other joints than these on one arm.
iCub has to set its arm joint angles (DoF) so that the hand is as closer to the object as possible (it reaches the object).
We use the Reinforcement Learning method of Machine learning (actually its Actor-Critic Learning Automaton specialization) to train iCub reaching objects.
Articles to read:
- Learning to reach by reinforcement learning using a receptive field based function approximation approach with continuous actions (2007, Minija Tamosiunaite, Tamim Asfour and Florentin Wörgötter)
- Aproximácia motorického priestoru ramena simulovaného robota (2010, Richard Korenčiak)
- Reinforcement Learning: An Introduction (2005, Richard S. Sutton and Andrew G. Barto)
Updated on Jaunuary 21, 2010
I have found some C++ examples, tutorials and toolboxes for Reinforcement Learning.
February 2020 M T W T F S S « Mar 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29
iCub on Twitter