Resolving perceptual aliasing in the presence of noisy sensors Conference Paper uri icon

abstract

  • Abstract Agents learning to act in a partially observable domain may need to overcome the problem of perceptual aliasing–ie, different states that appear similar but require different responses. This problem is exacerbated when the agent's sensors are noisy, ie, sensors may produce different observations in the same state. We show that many well-known reinforcement learning methods designed to deal with perceptual aliasing, such as Utile Suffix Memory, finite size history windows, eligibility traces, and memory bits, do not …

publication date

  • January 1, 2005