Compatible natural gradient policy search

Trust-region methods have yielded state-of-the-art results in policy search. A common approach is to use KL-divergence to bound the region of trust resulting in a natural gradient policy update. We show that the natural gradient and trust region optimization are equivalent if we use the natural parameterization of a standard exponential policy distribution in combination with compatible value function approximation. Moreover, we show that standard natural gradient updates may reduce the entropy of the policy according to a wrong schedule leading to premature convergence. To control entropy reduction we introduce a new policy search method called compatible policy search (COPOS) which bounds entropy loss. The experimental results show that COPOS yields state-of-the-art results in challenging continuous control tasks and in discrete partially observable tasks.

[1]  Jan Peters,et al.  Model-Free Trajectory-based Policy Optimization with Monotonic Improvement , 2016, J. Mach. Learn. Res..

[2]  Koray Kavukcuoglu,et al.  PGQ: Combining policy gradient and Q-learning , 2016, ArXiv.

[3]  Stefan Schaal,et al.  Natural Actor-Critic , 2003, Neurocomputing.

[4]  Sergey Levine,et al.  Trust Region Policy Optimization , 2015, ICML.

[5]  Hany Abdulsamad,et al.  Model-Free Trajectory Optimization for Reinforcement Learning , 2016, ICML.

[6]  Alex Graves,et al.  Asynchronous Methods for Deep Reinforcement Learning , 2016, ICML.

[7]  Guy Lever,et al.  Deterministic Policy Gradient Algorithms , 2014, ICML.

[8]  Stephen P. Boyd,et al.  Convex Optimization , 2004, Algorithms and Theory of Computation Handbook.

[9]  Jan Peters,et al.  Noname manuscript No. (will be inserted by the editor) Policy Search for Motor Primitives in Robotics , 2022 .

[10]  Tapani Raiko,et al.  International Conference on Learning Representations (ICLR) , 2016 .

[11]  Elman Mansimov,et al.  Scalable trust-region method for deep reinforcement learning using Kronecker-factored approximation , 2017, NIPS.

[12]  Alec Radford,et al.  Proximal Policy Optimization Algorithms , 2017, ArXiv.

[13]  Jeff G. Schneider,et al.  Covariant policy search , 2003, IJCAI 2003.

[14]  Germán Ros,et al.  CARLA: An Open Urban Driving Simulator , 2017, CoRL.

[15]  Yuval Tassa,et al.  Maximum a Posteriori Policy Optimisation , 2018, ICLR.

[16]  Yuval Tassa,et al.  Continuous control with deep reinforcement learning , 2015, ICLR.

[17]  Luís Paulo Reis,et al.  Model-Based Relative Entropy Stochastic Search , 2016, NIPS.

[18]  Matthieu Geist,et al.  Revisiting Natural Actor-Critics with Value Function Approximation , 2010, MDAI.

[19]  Pieter Abbeel,et al.  Benchmarking Deep Reinforcement Learning for Continuous Control , 2016, ICML.

[20]  Masashi Sugiyama,et al.  Guide Actor-Critic for Continuous Control , 2017, ICLR.

[21]  Jan Peters,et al.  Hierarchical Relative Entropy Policy Search , 2014, AISTATS.

[22]  Tom Schaul,et al.  Natural Evolution Strategies , 2008, 2008 IEEE Congress on Evolutionary Computation (IEEE World Congress on Computational Intelligence).

[23]  Joelle Pineau,et al.  Online Planning Algorithms for POMDPs , 2008, J. Artif. Intell. Res..

[24]  Nikolaus Hansen,et al.  Completely Derandomized Self-Adaptation in Evolution Strategies , 2001, Evolutionary Computation.

[25]  Shun-ichi Amari,et al.  Natural Gradient Works Efficiently in Learning , 1998, Neural Computation.

[26]  Yasemin Altun,et al.  Relative Entropy Policy Search , 2010 .

[27]  R. Rubinstein.  The Cross-Entropy Method for Combinatorial and Continuous Optimization , 1999 .

[28]  Yishay Mansour,et al.  Policy Gradient Methods for Reinforcement Learning with Function Approximation , 1999, NIPS.

[29]  François Laviolette,et al.  Domain-Adversarial Training of Neural Networks , 2015, J. Mach. Learn. Res..

[30]  Ronald J. Williams,et al.  Simple Statistical Gradient-Following Algorithms for Connectionist Reinforcement Learning , 2004, Machine Learning.

[31]  Sham M. Kakade,et al.  A Natural Policy Gradient , 2001, NIPS.

[32]  Shane Legg,et al.  Human-level control through deep reinforcement learning , 2015, Nature.

[33]  Guillaume Hennequin,et al.  Exact natural gradient in deep linear networks and its application to the nonlinear case , 2018, NeurIPS.

[34]  Jan Peters,et al.  Policy Search for Motor Primitives in Robotics , 2008, NIPS 2008.