PhD Defense by Emmanuel Boidot

Event Details
  • Date/Time:
    • Monday November 6, 2017
      3:00 pm - 5:00 pm
  • Location: TSRB 509
  • Phone:
  • URL:
  • Email:
  • Fee(s):
    N/A
  • Extras:
Contact
No contact information submitted.
Summaries

Summary Sentence: Ambush Games in Discrete and Continuous Environments

Full Summary: No summary paragraph submitted.

Ph.D. Thesis Defense by

 

Emmanuel Boidot

(Advisor: Prof. Eric Feron)

 

“Ambush Games in Discrete and Continuous Environments”

 

Monday, November 6, 2017 @ 3p.m.

TSRB 509

 

Abstract:

We consider an autonomous navigation problem, whereby a traveler aims at traversing an environment in which an adversary sets an ambush. A two-players zero- sum game is introduced, describing the initial strategy of the traveler and the ambusher based on a description of the environment and the traveler initial location and desired goal. The process is single-step in the sense that agents do not reevaluate their strategy after the traveler has started moving. Players’ strategies are computed as probabilistic path distributions, a realization of which is the path chosen by the traveler and the ambush location chosen by the ambusher.

 

A parallel is drawn between the discrete problem, where the traveler moves on a network, and the continuous problem, where the traveler moves in a compact subset of R2. Analytical optimal policies are derived. Assumptions from the Minimal Cut - Maximal Flow literature for continuous domains are used. The optimal value of the game is related to the maximum flow on the environment for sub-classes of games where the reward function for the ambusher is uniform. This proof is detailed in the discrete and continuous setups.

 

In order to relax the assumptions for the computation of the players’ optimal strategies, a sampling-based approach is proposed, inspired by recent sampling-based motion planning techniques. Given a uniform reward function for the ambusher, optimal strategies of the sampled ambush game are proven to converge to the optimal strategy of the continuous ambush game under some sampling and connectivity constraints. A linear program is introduced that allows for the computation of optimal policies. The sampling-based approach is more general in the sense that it is compatible with constrained motion primitives for the traveler and non-uniform reward functions for the ambusher.

The sampling-based game is used to create example applications for situations where no analytic solutions of the Continuous Ambush Game have been identified. This leads to more interesting games, applicable to real world robots using modern motion planning algorithms. Examples of such games are setups where the traveler’s motion satisfies Dubins’ kinematic constraints and setups where the reach of the ambusher is dependent on the speed of the traveler.

 

Committee Members:

Prof. Eric Feron (Advisor)

Prof. Panagiotis Tsiotras (AE)

Prof. John-Paul Clarke (AE)

Prof. Sam Coogan (ECE)

Prof. Eric Johnson (Penn State AE)

Additional Information

In Campus Calendar
No
Groups

Graduate Studies

Invited Audience
Faculty/Staff, Public, Graduate students, Undergraduate students
Categories
Other/Miscellaneous
Keywords
Phd Defense
Status
  • Created By: Tatianna Richardson
  • Workflow Status: Published
  • Created On: Oct 31, 2017 - 3:35pm
  • Last Updated: Oct 31, 2017 - 3:35pm