Corpus ID: 17333616

Information Theoretically Aided Reinforcement Learning for Embodied Agents

  title={Information Theoretically Aided Reinforcement Learning for Embodied Agents},
  author={Guido Mont{\'u}far and K. Zahedi and N. Ay},
  • Guido Montúfar, K. Zahedi, N. Ay
  • Published 2016
  • Computer Science, Mathematics
  • ArXiv
  • Reinforcement learning for embodied agents is a challenging problem. The accumulated reward to be optimized is often a very rugged function, and gradient methods are impaired by many local optimizers. We demonstrate, in an experimental setting, that incorporating an intrinsic reward can smoothen the optimization landscape while preserving the global optimizers of interest. We show that policy gradient optimization for locomotion in a complex morphology is significantly improved when… CONTINUE READING
    8 Citations
    A Unified Bellman Optimality Principle Combining Reward Maximization and Empowerment
    • 6
    • PDF
    VIME: Variational Information Maximizing Exploration
    • 395
    • PDF
    Adaptive Reward-Free Exploration
    • 4
    • PDF
    Curiosity-Driven Reinforcement Learning with Homeostatic Regulation
    • 16
    • PDF
    A unified strategy for implementing curiosity and empowerment driven reinforcement learning
    • 8
    • PDF
    Mega-Reward: Achieving Human-Level Play without Extrinsic Rewards
    • 5
    • PDF
    Factorized Mutual Information Maximization
    • 3
    • PDF


    Intrinsically Motivated Reinforcement Learning
    • 608
    • PDF
    Curiosity driven reinforcement learning for motion planning on humanoids
    • 59
    • PDF
    Learning and exploration in action-perception loops
    • 76
    • PDF
    Linear combination of one-step predictive information with an external reward in an episodic policy gradient setting: a critical analysis
    • 10
    • PDF
    An information-theoretic approach to curiosity-driven reinforcement learning
    • 126
    • PDF
    Reinforcement Learning: An Introduction
    • 27,047
    • PDF
    Intrinsically Motivated Learning in Natural and Artificial Systems
    • 204
    Curriculum learning
    • 2,208
    • PDF
    Higher Coordination With Less Control—A Result of Information Maximization in the Sensorimotor Loop
    • 65
    • PDF