Differences

This shows you the differences between two versions of the page.

--- projects [2017/01/05 21:10] – roumani
+++ projects [2017/01/05 21:15] (current) – roumani
@@ Line 5: / Line 5: @@
 **Supervisors**: Professors Michael Jenkin and Patrick Dymond
-Project: How should a group of autonomous agents moving within multiple locations choose and maintain a leader to coordinate and control them?
+**Project**: How should a group of autonomous agents moving within multiple locations choose and maintain a leader to coordinate and control them?
 Utilizing the concept of an infection algorithm — a process much like the spread of a disease in which agents infect each other with
 information — it is possible to develop a probabilistic approach to this kind of leader election problem. Solutions to this type of problem finds wide application in distributed computing, and in particular distributed computing of autonomous agents and sensors which must compute information about control structures with limited information about the presence/absence of potential leaders in the environment.
@@ Line 12: / Line 12: @@
 If time permits, the project will also study possible  enhancements of the infection algorithm as well as develop better upper bounds for the problem.
-Required skills: knowledge of Java or Python. Interest in algorithms for a distributed collection of simple robot agents. Completion of 3rd year courses in computer science or computer engineering.
+**Required skills**: knowledge of Java or Python. Interest in algorithms for a distributed collection of simple robot agents. Completion of 3rd year courses in computer science or computer engineering.
+======Asynchronicity in infection algorithms======
+**Supervisors**: Professors Michael Jenkin and Patrick Dymond
+**Project**: How does changing the model of synchronization changing run time bounds on infection algorithms?
+Infection algorithms are a class of algorithms within which individual agents exchange information via infection. That is, the algorithm proceeds by the various agents transmitting (infecting) each other with information. Under an assumption of synchronization — that is, a model in which no two agents can infect each other at precisely the same time — it is possible to derive models of expected time until all agents have been infected. But how does this algorithm adapt when agents can actually infect each other simultaneously?  This project will explore this problem. First, a simple simulation algorithm will be implemented to test infection rates when it is assumed that at a given time instant only one infection can occur. This algorithm will then be generalized to a model under which within a given time interval more than one infection can occur. Experimental validation will explore how existing infection algorithms perform under this more realistic model.  Simulations will be supported using a collection of real devices (Android devices) who can communicate with each other in the local environment using bluetooth and/or WIFI.
+**Required skills**: knowledge of Java or Python. Interest in algorithms for a distributed collection of simple robot agents. Completion of 3rd year courses in computer science or computer engineering.
+\\
+======Simultaneous localization and mapping (SLAM) aided by a single unique directional landmark======
+**Supervisors**: Professors Michael Jenkin and Patrick Dymond
+**Project**: How can a simple SLAM algorithm be improved if it is known a priori that there exists in the environment a single unique landmark that provides both position and orientation information?
+SLAM algorithms have been developed for a large number of different environments, and for relatively straightforward environments such as those found indoors, the problem can in many ways be considered solved. However, current algorithms have in the main concentrated on a general version of the problem in which no a priori information is assumed about the environment. From theoretical results we know that a single directional landmark is sufficient to solve the SLAM problem algorithmically, but can we exploit this result within the traditional probabilistic framework found in SLAM algorithms? Starting with a standard vision-based SLAM algorithm, this project will explore how a single landmark can be used to drive exploration and critically, drive algorithmic loop closure.
+**Required skills**: knowledge of Java or Python. Interest in algorithms for robots. Completion of 3rd year courses in computer science or computer engineering.