Transcript Slides 2013
Intelligent Patrolling Sarit Kraus Department of Computer Science Bar-Ilan University Collaborators: Noa Agmon, Gal Kaminka, Efrat Sless 1 Physical Security with Bounded Resources • Limited security resources prevent full security coverage at all times; allows adversaries to observe and exploit patterns in selective patrolling or monitoring. • Randomized patrolling or monitoring is necessary. 2 Randomized Checkpoints 3 Multi-Robot Adversarial Patrol • Motivation: • High security facilities • Large military bases • Neighborhood watch 4 Protecting a Parade: route is announced 5 Protecting a Moving Target: randomizing the route 6 Why Do We Need Automated Methods for Randomization? • People are not good at randomization • The randomized strategy should depend on: o the adversary and the defenders’ utilities o the environment parameters o the defender’s resources. 7 Multi-Robot Adversarial Patrol: The Environment • Perimeter o Divided into segments o Uniform time-distance o Robot travels through one segment per time unit • Adversary o Tries to penetrate through the perimeter o Takes t>0 time units to penetrate 8 Robotic Model • k homogenous robots • Robotic movement model: o Robots’ movement is directed o Turning around “costs” τ time units 9 Algorithm Framework • Patrol algorithm: o Continue straight with probability p o Turn around with probability 1-p • PPD: Probability of Penetration Detection • p depends on the distance between robots and on the penetration time. 10 Robot Movement • Optimal: Robots are uniformly placed around the perimeter • Coordinated o If decide to turn around they do it simultaneously o Preserve uniform distance 11 Robot Movement • Optimal: Robots are uniformly placed around the perimeter • Coordinated – If decide to turn around they do it simultaneously – Preserve uniform distance 12 Robot Movement • Optimal: Robots are uniformly placed around the perimeter • Coordinated – If decide to turn around they do it simultaneously – Preserve uniform distance 13 Robot Movement • Optimal: Robots are uniformly placed around the perimeter • Coordinated – If decide to turn around they do it simultaneously – Preserve uniform distance 14 Handling Events What if a robot needs to inspect the penetration? • Once a penetration is detected, one robot is extracted from the team to inspect it • Coordinated attacks are beneficial to the adversary 15 What Happens If Penetration Detected? 16 What Happens If Penetration Detected? • Robot that detected the penetration will inspect it • Other k-1 robots will spread uniformly o To achieve optimal behavior for k-1 robots Phase 1: k robots (before event), steady state Phase 2: Reorganization Phase 3: k-1 robots (after event), steady state Optimal patrol known 17 Naïve Approach • Deterministic: Each robot goes straight to its final position 18 Randomized Reorganization • Challenges: o each robot needs to move differently o How much time to spend on the reorganization? • We randomized over possible paths • Finding the strategy is complex, in theory, but we used heuristics to find it in reasonable time 19 Physical Security with Bounded Resources: Summary • Randomized patrolling or monitoring is necessary. • Automated randomization is important • Interesting problems? [email protected] 20