We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations

DBLP - CS Bibliography


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Artificial Intelligence

Title: Information Theoretically Aided Reinforcement Learning for Embodied Agents

Abstract: 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 supplementing the extrinsic reward by an intrinsic reward defined in terms of the mutual information of time consecutive sensor readings.
Comments: 10 pages, 4 figures, 8 pages appendix
Subjects: Artificial Intelligence (cs.AI); Robotics (cs.RO); Optimization and Control (math.OC); Machine Learning (stat.ML)
MSC classes: 68T05, 68T40
Cite as: arXiv:1605.09735 [cs.AI]
  (or arXiv:1605.09735v1 [cs.AI] for this version)

Submission history

From: Guido F. Montufar [view email]
[v1] Tue, 31 May 2016 17:30:54 GMT (2680kb,D)

Link back to: arXiv, form interface, contact.