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: Adaptive Information Belief Space Planning

Abstract: Reasoning about uncertainty is vital in many real-life autonomous systems. However, current state-of-the-art planning algorithms cannot either reason about uncertainty explicitly, or do so with a high computational burden. Here, we focus on making informed decisions efficiently, using reward functions that explicitly deal with uncertainty. We formulate an approximation, namely an abstract observation model, that uses an aggregation scheme to alleviate computational costs. We derive bounds on the expected information-theoretic reward function and, as a consequence, on the value function. We then propose a method to refine aggregation to achieve identical action selection with a fraction of the computational time.
Subjects: Artificial Intelligence (cs.AI); Robotics (cs.RO)
Cite as: arXiv:2201.05673 [cs.AI]
  (or arXiv:2201.05673v2 [cs.AI] for this version)

Submission history

From: Moran Barenboim [view email]
[v1] Fri, 14 Jan 2022 21:12:00 GMT (1309kb,D)
[v2] Fri, 28 Jan 2022 07:43:35 GMT (1312kb,D)

Link back to: arXiv, form interface, contact.