Download presentation
Presentation is loading. Please wait.
1
Spider Search: an Efficient and Non-Frontier-Based Real-Time Search Algorithm Presenter: Chao Lin Chu 599B Advisor: Dr. Russell J. Abbott
2
Outline ● Introduction ● Problem formulation ● AI approaches ● Experimental result ● Future work ● Conclusion
3
Introduction ● Real-time searches use limited depth. ● Standard searches don't work on large spaces. ● We can't simply reduce the depth to solve problem. ● Spider Search uses a very low constant space and time to solve this.
4
Problem Formulation – Boat/Torpedo problem ● 2x2 square sea, toroidal world, from -1 to 1 ● Torpedoes are faster. Boat has tighter turning radius. ● The smart boat plans ahead to avoid collision. ● They are small compared to the space. ● Small lookahead is not enough.
5
AI Approaches ● Breadth-First Search ● Depth-First Search ● A* Search ● Learning Real-Time A* ( LRTA* ) ● Spider Search
6
Spider Search ● Different from Frontier-based Searches ● Separate Generation Selection
7
Spider Search ● Population and fitness from Genetic Algorithm. ● Use probabilities to control randomness of path distribution. ● Limited depth. ● Keep the best path for next step.
8
Spider Search ● Turn the binary tree into an N-ary tree. LLLLL RLLLL RRLLL RRRLL RRRRL RRRRR ● Build out significantly longer paths. ● Standard AI searches don't do this.
10
Fitness Function
11
Spider Search ● Then pick the best leaf node. ● The next state will be the neighbor of the root that is on the path to the best leaf node. ● Then keep the best path for the next step.
12
Result – part 1
13
Result – part 2
15
Comparison Between LRTA* and Spider Search ● LRTA* -- grow exponentially with depth ● Spider Search: O( d * ( d + 1 ) ) where d is the depth. O( c d ) where c is a constant of population ● The most important thing is that the depth is not changed in a single run so the required time and space is a small constant throughout each run, for instance, 75 * 14 = 1050.
16
Future Work ● Enhancement of the Boat ● Enhancement of the Torpedoes ● Adapt Spider Search into Other Search Algorithm
17
Conclusion ● Not optimal but efficient. ● Use very low constant space and time.
18
Thank you!
19
Spider Search
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.