Optimal Direct Policy Search

  • Tobias Glasmachers
  • Jürgen Schmidhuber
Part of the Lecture Notes in Computer Science book series (LNCS, volume 6830)


Hutter’s optimal universal but incomputable AIXI agent models the environment as an initially unknown probability distribution-computing program. Once the latter is found through (incomputable) exhaustive search, classical planning yields an optimal policy. Here we reverse the roles of agent and environment by assuming a computable optimal policy realizable as a program mapping histories to actions. This assumption is powerful for two reasons: (1) The environment need not be probabilistically computable, which allows for dealing with truly stochastic environments, (2) All candidate policies are computable. In stochastic settings, our novel method Optimal Direct Policy Search (ODPS) identifies the best policy by direct universal search in the space of all possible computable policies. Unlike AIXI, it is computable, model-free, and does not require planning. We show that ODPS is optimal in the sense that its reward converges to the reward of the optimal policy in a very broad class of partially observable stochastic environments.


Optimal Policy Reinforcement Learning Turing Machine Markov Decision Process Direct Policy 
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.


Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.


  1. 1.
    Hutter, M.: Universal Artificial Intelligence: Sequential Decisions based on Algorithmic Probability. Springer, Berlin (2004)Google Scholar
  2. 2.
    Levin, L.: Universal sequential search problems. Problems of Information Transmission 9(3), 265–266 (1973)Google Scholar
  3. 3.
    Schaul, T., Schmidhuber, J.: Towards Practical Universal Search. In: Proceedings of the Third Conference on Artificial General Intelligence, Lugano (2010)Google Scholar
  4. 4.
    Schmidhuber, J.: Sequential decision making based on direct search (Lecture Notes on AI 1828). In: Sun, R., Giles, C.L. (eds.) IJCAI-WS 1999. LNCS (LNAI), vol. 1828, p. 213. Springer, Heidelberg (2001)CrossRefGoogle Scholar
  5. 5.
    Schmidhuber, J.: Optimal Ordered Problem Solver. Machine Learning 54, 211–254 (2004)zbMATHCrossRefGoogle Scholar
  6. 6.
    Schmidhuber, J.: Gödel machines: Fully Self-Referential Optimal Universal Self-Improvers. In: Goertzel, B., Pennachin, C. (eds.) Artificial General Intelligence, pp. 119–226 (2006)Google Scholar
  7. 7.
    Schmidhuber, J.: Ultimate Cognition à la Gödel. Cognitive Computation 1(2), 177–193 (2009)CrossRefGoogle Scholar
  8. 8.
    Schultz, W., Dayan, P., Montague, P.R.: A neural substrate of prediction and reward. Science 275(5306), 1593 (1997)CrossRefGoogle Scholar
  9. 9.
    Veness, J., Ng, K.S., Hutter, M., Silver, D.: A Monte Carlo AIXI Approximation. Technical Report 0909.0801, arXiv (2009)Google Scholar

Copyright information

© Springer-Verlag Berlin Heidelberg 2011

Authors and Affiliations

  • Tobias Glasmachers
    • 1
  • Jürgen Schmidhuber
    • 1
  1. 1.IDSIAUniversity of LuganoManno-LuganoSwitzerland

Personalised recommendations