Αλμπέρτο Ερσπάμερ

Envelope-Based Approaches to Real-Time Heuristic Search

Transcript

1. Envelope-Based Approaches to Real-Time Heuristic Search Kevin C. Gall, Bence Cserna, Wheeler Ruml Almperto Erspamer Master program “Computer science: Data,Information and Knowledge Management”, Athens, NKUA, 2020
2. A* Search A* selects the path that minimizes f(s) = g(s)+h(s) g(s) h(s) where s is the next node on the path, g(s) is the cost of the path from the start node to s, and h(s) is a heuristic function that estimates the cost of the cheapest path from s to the goal At each iteration of its main loop, A* needs to determine which of its paths to extend..
3. TBA* Search If Traceback terminates at agent’s or root position best node founded is defined as the new target If the agent is on the path of the target continue to that Path. Otherwise, move back to a previous state Time bounded A*… 2 1 3 4
4. TBA* Variants • Time- Bounded A* : Weighting, Greedy variants • WA* Variant: f(s) = g(s)+w∗h(s) • BFS Variant: Omit g(s) “There is no reason that the agent must necessarily find an optimal path from the root to a goal, as its actual trajectory is likely to be suboptimal.”
5. TBA* Variants: Problems • TB(WA*) challenge • TB(BFS) challenge
6. Intra-Envelope Search …a new Envelope Search approach to achieve better “shortcuts” through the state space towards a target node
7. Intra-Envelope Search • Bounded-Exploration and Path extraction • Backward search similar with traceback in TBA* • Goal of backward search is set the instead of agent. ,1 ,2 Always choose best
8. Intra-Envelope Search-Highlights • By anchoring neither the search nor the path extraction in the increasingly less relevant root node, I-ES is very effective at navigating dense graphs via shortcuts when new target nodes are selected. • I-ES spends computation on its extra intra-envelope search instead of continuing to expand the frontier with all available time
9. Intra-Envelope Search vs TBA* • I-ES exhibits superior performance in highly-connected domains like grid pathfinding where shortcuts were easily extracted, although it did not provide any advantage in more sparsely- connected state spaces such as racetrack.
10. References • A* search algorithm – Wikipedia • Time-Bounded Best-First Search for Reversible and Non-reversible Search Graphs, C. Hernandez, J.A.Baier and R.Asin, Journal of Artificial Intelligence Research, 2016 • TBA*: Time-Bounded A*, Y.Bjornsson, V.Bulitko and N.Sturtevant, Proceedings of the 21st International Joint Conference on Artificial Intelligence, Pasadena, California, USA, July 11-17, 2009 • A-star search algorithm: A-Star Search algorithm with step-by-step code, B.Roy, 2019
11. Thank you for your attention