Examining D and D Lite

1
Examining D and D Lite

• Robot path planning in partially known and
  dynamic environments
• Dahl Groenke

2
Goal

• To find out what the differences are
• To determine how closely the algorithms are
  related
• Overall question Are D and D Lite different
  optimizations of the same base algorithm?

3
Focussed D

• D developed by Anthony Stentz in 1994
• Focussed D developed in 1995
• A widely used algorithm for robot navigation
• Never extended

4
Ds Strategy

• Each state (vertex) has a back pointer that
  points toward the optimal goal path
• A priority queue contains states that need to be
  updated
• The key value for each state determines ordering,
  k(X) minumum of all goal distance estimates
• Any neighbors of the current state that should
  point back to the current state according to goal
  distance estimates are changed and added to the
  queue

5
Focusing D

• The original D did not take into account the
  robots position when expanding vertices
• Stentz modified the ordering of the priority
  queue by incorporating both the distance to the
  goal and the distance to the robot.

6
D
7
Focussed D
8
(No Transcript)
9
D Lite

• Developed by Koenig and Likhachev
• Based on Lifelong Planning A
• Initial runs are identical to A
• Subsequent runs are incremental A

10
D Lites Strategy

• Each vertex s has a g(s), start distance
  estimate, and a rhs(s), one step lookahead start
  distance estimate
• A priority queue hold states to be processed,
  ordered according to start distance and goal
  distance

11
(No Transcript)
12
Similarities

• Similarities
• Search from goal to start
• Use combination of start and goal heuristics
• Maintain global policy instead of one path
• Remove local inconsistencies to build path

13
D Lites superior performance

• D has
• 30 - 50 more vertex accesses
• 15 - 40 more vertex expansions
• 80 - 100 more heap percolates

14
Reasons for Advantage

• D Lite use of lookahead values could lead to
  more focused propagation of change