Applying the Episodic Natural Actor-Critic Architecture to Motor Primitive Learning

By Jan Peters, Stefan Schaal. 2007

30 citations recorded by the source corpus.

Subfields: Reinforcement Learning, Robotics & Control

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Reinforcement Learning

In Robotics & Control

Citations in the corpus

Reset-free guided policy search: Efficient deep reinforcement learning with stochastic initial states (2016, methodology): “We compare our algorithm to three prior RL methods, TRPO [6], RWR [20], and REPS [1], as well as two prior GPS methods, including on-policy MDGPS [16] and an older formulation based on the Bregman Alternating Direction Method of Multiplers (BADMM) [5].”
Online Multi-Task Learning for Policy Gradient Methods (2014, background): “…of state-action pairs forms a trajectory τ = [x 0:H , a 0:H ] over a possibly infinite horizon H. Policy gradient methods (Sutton et al., 1999; Peters & Schaal, 2008; Peters & Bagnell, 2010) have shown success in solving high-dimensional problems, such as robotic control (Peters & Schaal, 2007).”
A Survey on Policy Search for Robotics (2013, methodology): “We distinguish between two EM-update procedures, called Monte-Carlo (MC)-EM approaches [39, 60] and Variational Inference for Policy Search [49], which we will discuss in the following sections.”
Learning deep neural network policies with continuous memory states (2015, methodology): “Besides guided policy search, we evaluated the memory states approach with the reward-weighted regression (RWR) algorithm [23], [24].”
Policy Improvement : Between Black-Box Optimization and Episodic Reinforcement Learning (2013, methodology): “Since these algorithms have already been covered in extensive surveys (Peters & Schaal, 2008a, 2007; Kober & Peters, 2011), we do not present them in full detail here, as this does not serve the particular aim of this section.”
Natural Actor-Critic (2003, background): “Additional to our work presented at ESANN 2007 in [19] and its earlier, preliminary versions (see e.”
Learning Complex Neural Network Policies with Trajectory Optimization (2014, methodology): “We also compared to cost-weighted regression, which fits the policy to previous on-policy samples weighted by the exponential of their reward (negative cost) (Peters & Schaal, 2007; Kober & Peters, 2009).”