BEGIN:VCALENDAR

VERSION:2.0

PRODID:-//wp-events-plugin.com//6.6.4.4//EN

TZID:Asia/Jerusalem

X-WR-TIMEZONE:Asia/Jerusalem
BEGIN:VEVENT

UID:0-435@aerospace.technion.ac.il

DTSTART;TZID=Asia/Jerusalem:20170306T163000

DTEND;TZID=Asia/Jerusalem:20170306T173000

DTSTAMP:20230530T181048Z

URL:https://aerospace.technion.ac.il/events/efficient-belief-space-panning
 -in-high-dimensional-state-spaces-by-exploiting-sparsity-and-calculation-r
 e-use/

SUMMARY:Efficient Belief Space Planning in High-Dimensional State Spaces by
  Exploiting Sparsity and Calculation Re-use
DESCRIPTION:Lecturer:Dmitry Kopitkov\n Faculty:Technion Autonomous Systems 
 Program (TASP)\n Institute:Technion – Israel Institute of Technology\n L
 ocation:Classroom 165\, ground floor\, Library\, Aerospace Eng.\n Zoom: \n
  Abstract: \n Details: \n Belief space planning (BSP) is a fundamental pro
 blem in robotics and artificial intelligence\, with applications including
  autonomous driving\, sensor deployment and active SLAM. The goal is to au
 tonomously determine best actions according to a specified objective funct
 ion\, given the current belief about random variables of interest (e.g. ro
 bot poses\, tracked target or mapped environment)\, while accounting for d
 ifferent sources of uncertainty. The objective function typically contains
  multiple terms\, such as distance to goal\, control cost and a final beli
 ef uncertainty. The latter measures the amount of information contributed 
 by candidate actions and is the most computationally expensive term to eva
 luate.\nI present an efficient approach for evaluating the information the
 oretic term within BSP\, where during belief propagation the state vector 
 can stay fixed in size or be augmented by additional variables (e.g. new r
 obot poses). Both unfocused and focused problem settings are considered\, 
 whereas uncertainty reduction of the entire system or only of chosen varia
 bles is of interest\, respectively. State of the art approaches typically 
 propagate the belief state\, for each candidate action\, through calculati
 on of the posterior information (or covariance) matrix and subsequently co
 mpute its determinant (required for entropy). In contrast\, our approach r
 educes run-time complexity by avoiding these calculations. We formulate th
 e problem in terms of factor graphs and show that belief propagation is no
 t needed\, requiring instead a one-time calculation that depends on state 
 dimensionality\, and per-candidate calculations that are independent of th
 e latter. To that end\, we develop an augmented version of the matrix dete
 rminant lemma\, and show computations can be re-used when evaluating impac
 t of different candidate actions. These two key ingredients result in a co
 mputationally efficient (augmented) BSP approach that accounts for differe
 nt sources of uncertainty and can be used with various sensing modalities.
CATEGORIES:Seminars
LOCATION:Classroom 165\, ground floor\, Library\, Aerospace Eng.

END:VEVENT

BEGIN:VTIMEZONE

TZID:Asia/Jerusalem

X-LIC-LOCATION:Asia/Jerusalem

BEGIN:STANDARD

DTSTART:20161030T010000

TZOFFSETFROM:+0300

TZOFFSETTO:+0200

TZNAME:IST

END:STANDARD

END:VTIMEZONE
END:VCALENDAR