mukhes3/multi-agent-pursuit-in-urban-evironment

This is a course project for the 'Distributed Systems & Sensors' class I had taken at RPI

Multi-agent-pursuit-in-urban-evironment

This is a course project for the 'Distributed Systems & Sensors' class I had taken at RPI. We called this project 'Catching Batman'.

Introduction

Consider an urban environment where there are many agents that move along a grid-like network of roads to travel from some source location to some destina- tion location. These cars are constrained to obey traffc laws, that is, their speed is bounded by the speed limit, and they can only travel through intersections In addition, there is a static network of sensors at each intersection that record when a car passes through and what its velocity is at that time. Our problem is that of pursuing a rogue agent traveling through the city with a set of drone vehicles, which can be thought of as the police. These drones can communicate with the stationary network and can be treated as mobile nodes within the net- work. We assume that the stationary nodes are also capable of distinguishing the rogue agent from the rest of the vehicles, and can pass this information to the pursuing agents. The pursuing agents are further constrained to maintain safety, that is, they will maintain some safe distance from other vehicles and will not unsafely pass through red lights, whereas the rogue agent is not required to follow these constraints.

In order to address the problem, we have the stationary network collect data concerning the current location and movement direction of the rogue agent, as well as current congestion levels on different sections of road, and make this information available to the pursuing agents. A wireless sensor network demands implementation of protocols and algo- rithms that make effective use of limited resources and counter the challenges posed by their operating environment. The network in our case comprises many static nodes and a few mobile nodes. The major limitation was the limited battery power for the stationary nodes. In wireless sensor networks, communica- tions takes the most amount of energy, so it is very essential to strike the right balance between data processing and transmission of raw data. We assumed that sensor nodes have the capability to do some amount of signal processing before they forward or disseminate the data. The topology is such that one can expect that at a particular time only a few stationary sensor will be in the vicinity of a mobile sensor. This boosts the case for giving control on part of the mobile nodes to initiate and fetch information from the stationary nodes. The degree of mobility of the nodes is another factor that determined the effciency of pursuit algorithms used. We have tested our algorithms by simulating the system in MATLAB while varying different parameters of the simulation, such as the starting location of the agents, the number of pursuers, the density of vehicles, the communication speed and range of sensors, etc.

An example simulation is seen here: https://www.youtube.com/watch?v=HhI5zNiAhKU