Policy Gradient Methods for Robotics Export

@proceedings{citeulike:2361557, abstract = {The acquisition and improvement of motor skills and control policies for robotics from trial and error is of essential importance if robots should ever leave precisely pre-structured environments. However, to date only few existing reinforcement learning methods have been scaled into the domains of high-dimensional robots such as manipulator, legged or humanoid robots. Policy gradient methods remain one of the few exceptions and have found a variety of applications. Nevertheless, the application of such methods is not without peril if done in an uninformed manner. In this paper, we give an overview on learning with policy gradient methods for robotics with a strong focus on recent advances in the field. We outline previous applications to robotics and show how the most recently developed methods can significantly improve learning performance. Finally, we evaluate our most promising algorithm in the application of hitting a baseball with an anthropomorphic arm}, author = {Peters, Jan and Schaal, Stefan}, booktitle = {Intelligent Robots and Systems, 2006 IEEE/RSJ International Conference on}, citeulike-article-id = {2361557}, citeulike-linkout-0 = {http://dx.doi.org/10.1109/IROS.2006.282564}, citeulike-linkout-1 = {http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=4058714}, doi = {10.1109/IROS.2006.282564}, journal = {Intelligent Robots and Systems, 2006 IEEE/RSJ International Conference on}, keywords = {actor-critic, policy\_gradient}, pages = {2219--2225}, posted-at = {2008-02-11 07:13:30}, priority = {0}, title = {Policy Gradient Methods for Robotics}, url = {http://dx.doi.org/10.1109/IROS.2006.282564}, year = {2006} }

TY - CONF ID - citeulike:2361557 L3 - citeulike-article-id:2361557 N2 - The acquisition and improvement of motor skills and control policies for robotics from trial and error is of essential importance if robots should ever leave precisely pre-structured environments. However, to date only few existing reinforcement learning methods have been scaled into the domains of high-dimensional robots such as manipulator, legged or humanoid robots. Policy gradient methods remain one of the few exceptions and have found a variety of applications. Nevertheless, the application of such methods is not without peril if done in an uninformed manner. In this paper, we give an overview on learning with policy gradient methods for robotics with a strong focus on recent advances in the field. We outline previous applications to robotics and show how the most recently developed methods can significantly improve learning performance. Finally, we evaluate our most promising algorithm in the application of hitting a baseball with an anthropomorphic arm JF - Intelligent Robots and Systems, 2006 IEEE/RSJ International Conference on EP - 2225 TI - Policy Gradient Methods for Robotics T2 - Intelligent Robots and Systems, 2006 IEEE/RSJ International Conference on SP - 2219 KW - actor-critic KW - policy_gradient AU - Peters, Jan AU - Schaal, Stefan PY - 2006/// UR - http://dx.doi.org/10.1109/IROS.2006.282564 ER -