Path Planning in a Dynamic
Environment
Scott McKeever, Nora Szasz,
Dennis Gregorovic, Angel Chang
Amit Seshan
Introduction
• What is path planning and whom does it
concern?
– Agents in Games
– UAVs
– You!
• Unmanned Autonomous Vehicles (UAVs)
should perform “intelligently”
• One problem UAVs encounter is getting
from point A to B “safely and cheaply”
Overview
• Problem Description
• Network Description
• Algorithms
• Demonstration of Algorithms
• Conclusion
Problem Description
• Maneuver a vehicle from point A to point B, using
little to no a-priori information, with limited sensor
range, while in the presence of dynamic obstacles.
• Ojective: Find a path that is optimal according to
some cost function.
Network Description
• What is the Structure of the Network?
– Nodes are a transformation of the map
• Grid Idea (used in our demonstration)
• Beacon Approach
– Arcs
• General definition: (if node j is
neighbor(node k) then arc {k,j} exists)
• A neighbor of node x (or gridpoint x) is
usually the spatially adjacent gridpoints to x
Network Description (cont.)
• Arc Cost Metrics
– Can take on many forms
– Usually a function of fuel or battery usage
– Can be as simple as time to maneuver along
an arc, or the distance of the arc
– Arcs that are forbidden (due to obstacles)
are assigned an infinitely high cost
Algorithms
• Greedy Algorithms
• Static Case
– Dijkstra’s Shortest Path
– A* Algorithm
• Dynamic Case with limited visibility
– D* Algorithm
A* Pseudo-code
Begin
//find potentials in G
(i):=heuristic(i) i  N;
//preprocess G: generate G
do
cij

:= cij - (i) + (j) (i,j)  A
//find shortest reduced cost paths s  t
Dijkstra(G )
• Advantages
• Disadvantages
D* Pseudo-code
Begin
//initialize variables
t(i):=NEW i  N
d(s):= 0; t(s):=OPEN
S:={s}
while Xt;
//compute shortest path
while X  S do
PROCESS-STATE;
//shortest path {X} is now computed
proceed on {X} until obstacle is detected;
if obstacle is detected
MODIFY-COST;
end;
end;
Demonstration
• Background
– Gridded Map
– Agent can only move
left, right, up, or
down
– Varying Terrain
– Distance = Manhattan
Distance
• Dijkstra
• A*
• D*
• Greedy Algorithms
– Learning Algorithm
•Link to Demo
Future Challenges and
Obstacles
• Many Pieces to the
Puzzle are Still
Unsolved
– Smooth Trajectories
– Stochastic Nature of
Data
– Moving Target
Questions ?