## Overview

Important theoretical aspects of mechanisms for multi-robot task allocation have, to date, been largely ignored. In this project, we are trying to address part of this negligence by formally studying the problem within an organizational framework developed in the Operations Research community. In particular, we are currently exploring multi-robot task allocation as an instance of the well-known Optimal Assignment Problem. In this light, we have recently analyzed and compared the algorithmic characteristics of several existing approaches to the problem.

## Introduction

Since the early 1990s, the problem of task allocation in multi-robot systems has received significant and increasing interest in the research community. As researchers design, build, and use cooperative multi-robot systems, they invariably encounter the question: "which robot should execute which task?" This question must be answered, even for relatively simple multi-robot systems, and the importance of task allocation grows with the complexity, in size and capability, of the system under study. Even in the simplest case of homogeneous robots with fixed, identical roles, intelligent allocation of tasks is required for good system performance, if only to minimize physical interference.

Of course, task allocation need not be explicit; it may instead emerge from the interactions of the robots (physical and otherwise), as is the case with coordination methods put forward by proponents of swarm robotics. Such an emergent system, when constructed skillfully, can be extremely effective and may provide the simplest and most elegant solution to a problem. However, it is a solution to a {\em specific} problem, and if robots are to be generally useful, we believe that they must be capable of solving a variety of problems.

Over the years, a significant body of work has been done on explicit multi-robot task allocation (MRTA), generally involving task-oriented inter-robot communication. A variety of such architectures have been proposed; while they are sometimes evaluated experimentally, they are rarely subject to formal analysis. It is coordination architectures of this type on which we are currently focusing our analysis. Our aim is to address two key shortcomings of the MRTA work to date:

• computation and communication requirements are generally unknown
• aside from experimental validation in specific domains, there is no characterization of the solution quality that can be expected

## Problem Statement

We claim that multi-robot task allocation can be reduced to an instance of the Optimal Assignment Problem (OAP), a well-known problem from Operations Research. A recurring special case of particular interest in several fields of study, this problem can be formulated in many ways. Given our application domain, it is fitting to describe the problem in terms of jobs and workers. There are n workers, each looking for one job, and m available jobs, each requiring one worker. The jobs can be of different priorities, meaning that it is more important to fill some jobs than others. Each worker has a nonnegative skill rating estimating his/her performance for each potential job (if a worker is incapable of undertaking a job, then the worker is assigned a rating of zero for that job). The problem is to assign workers to jobs in order to maximize the overall expected performance, taking into account the priorities of the jobs and the skill ratings of the workers.

Our multi-robot task allocation problem can be posed as an assignment problem in the following way: given n robots, m prioritized (i.e., weighted) single-robot tasks, and estimates of how well each robot can be expected to perform each task, assign robots to tasks so as maximize overall expected performance. However, because the problem of task allocation is a dynamic decision problem that varies in time with phenomena including environmental changes, we cannot be content with this static assignment problem. Thus we complete our reduction by iteratively solving the static assignment problem over time.

Of course, the cost of running the assignment algorithm must be taken into account. At one extreme, a costless algorithm can be executed arbitrarily fast, ensuring an optimal assignment over time. At the other extreme, an expensive algorithm that can only be executed once will produce a static assignment that is only initially optimal and will degrade over time. Finally there is the question of how many tasks are considered for (re)assignment at each iteration. In order to create and maintain an optimal allocation, the assignment algorithm must consider (and potentially reassign) every task in the system. Such an inclusive approach can be computationally expensive and, indeed, some implemented approaches to MRTA use heuristics to determine a subset of tasks that will be considered in a particular iteration.

Together, the cost of the static algorithm, the frequency with which it is executed, and the manner in which tasks are considered for (re)assignment will determine the overall computational and communication overhead of the system, as well as the solution quality. Thus it is these characteristics of MRTA architectures in which we are interested and that we are currently analyzing.

## Utility

Utility is a unifying, if sometimes implicit, concept in economics, game theory, and operations research, as well as multi-robot coordination. The idea is that each individual can somehow internally estimate the value (or the cost) of executing an action. It is variously called fitness, valuation, and cost. Since the exact formulation varies from system to system, we now give an instructive and generic, yet practical, definition of utility for multi-robot systems.

We assume that each robot is capable of estimating its fitness for every task of which it is capable. This estimation includes two factors, both task- and robot-dependent:

• expected quality of task execution, given the method and equipment to be used (e.g., the accuracy of the map that will be produced using a laser range-finder)
• expected resource cost, given the spatio-temporal requirements of the task (e.g., the power that will be required to drive the motors and laser range-finder in order to map the building)
If these two quantities are defined on some standardized scale, then we can produce a non-negative combined utility estimate simply by subtracting cost from quality.

The robots' utility estimates will be inexact for a number of reasons, including sensor noise, general uncertainty, and environmental change. These unavoidable characteristics of the multi-robot domain will necessarily limit the efficiency with which coordination can be achieved. We treat this limit as exogenous, on the assumption that lower-level robot control has already been made as reliable, robust, and precise as possible and thus that we are incapable of improving it. When we later discuss "optimal" allocation solutions; we mean "optimal" in the sense that, given the union of all information available in the system (with the concomitant noise, uncertainty, and inaccuracy), it is impossible to construct a solution with higher overall utility.

Note that although we have not discussed planning or learning, our definition of utility permits their introduction. The addition of a predictive model, for example, will (presumably) improve the accuracy of utility estimates by considering expected future events. To achieve the best system performance, such techniques should be used whenever possible.

## Analysis

Having given a formal statement of the MRTA problem, we are now in a position to analyze some of the key task allocation architectures from the literature. We have recently analyzed six approaches to MRTA, focusing on three characteristics:

• computation requirements
• communication requirements

In part because of trends in the research community that stress the importance of experimental validation with physical robots, such theoretical aspects of multi-robot coordination mechanisms have been largely ignored. However, they are vitally important to the study, comparison, and objective evaluation of the mechanisms. The large-scale and long-term behavior of the system will be strongly determined by the fundamental characteristics of the underlying algorithm(s). Thus we endeavor to derive and explicate those characteristics here. Before we continue, however, it will be necessary to explain the methodology that we use in our analysis.

### Results

See the paper listed below.

## Relevant Publications

• "A formal analysis and taxonomy of task allocation in multi-robot systems".
Brian P. Gerkey and Maja J Mataric´. Intl. Journal of Robotics Research, 23(9):939-954, September 2004. (Also Technical Report CRES-03-013)
[gzipped PS: 116k][PDF: 176k]
• "Multi-Robot Task Allocation: Analyzing the Complexity and Optimality of Key Architectures".
Brian Gerkey and Maja J Mataric´. In Proceedings of the IEEE International Conference on Robotics and Automation (ICRA 2003), pages 3862-3868, Taipei, Taiwan, September 14 - 19, 2003. (Also Technical Report CRES-02-005)
[gzipped PS: 56k][PDF: 76k]
• "A Framework for Studying Multi-Robot Task Allocation".
Brian Gerkey and Maja J Mataric´. In A.C. Schultz and others, editors, Multi-Robot Systems: From Swarms to Intelligent Automata, Volume II, pages 15-26, Kluwer Academic Publishers, the Netherlands, 2003 (presented at 2003 International Workshop on Multi-Robot Systems).
[gzipped PS: 60k][PDF: 132k]

Last updated $Date: 2004/12/08 16:16:39$

## Learning Goal

Examine culture jamming artwork and create your own, using creative commons photos and Photoshop.

I got this idea from Mr. Arnett at St. Mary’s High School.

What is culture jamming?

According to Naomi Klein, taken from her book, No Logo, culture jamming is

the practice of parodying advertisements and hijacking billboards in order to drastically alter their messages.  Streets are public spaces, adbusters argue, and since most residents can’t afford to counter corporate messages by purchasing their own ads, they should have the right to talk back to images they never asked to see.

One of the more respected and founders of culture jamming is Rodriguez de Gerada.

Rodriguez was in a New York Art Group called Artfux.  They had specific targets when it came to culture jamming which he mentions on his website,

We decided to focus on the disproportionately high amount of damaging products (get drunk quick beverages and menthol cigarette brands) being advertised in poor and minority areas in the Tri State area. The billboard alterations from this time are among my favorites because there was a very clear goal and plan of action. We would illegally alter or replace a tobacco or alcohol ad with a new statement and image that spoke about the negative effects of these products. We would then send out press releases with photos of our exploits. We received a massive amount of attention.

The Billboard Liberation Front from San Francisco has been altering billboards since 1977.

Here are some of their more recent efforts:

This is the before image:

This is the after image:

Adbusters has probably been the poster organization when it comes to culture jamming.  They are based out of Vancouver and have an international circulation of over 120,000 for their magazine (Wikipedia).  They have been  behind the “Buy Nothing Day” campaign and helped promote “TV Turnoff Week”.  They are well known for their spoof ads which target large multinational corporations such as McDonald’s and Nike.  Click here to see some of their popular spoof ads.

Culture-jamming has become quite popular and American Labor rights activist explains why in Klein’s No Logo,

There are certain corporations which market themselves so aggressively, which are so intent on stamping their image on everybody and every street, that they build up a reservoir of resentment among thinking people.  People resent the destruction of culture and its replacement with these mass-produced corporate logos and slogans.  It represents a kind of cultural fascism.

Sao Paulo in Brazil decided to ban billboards years ago.
What does it look like now?
[Link to Flickr images of Sao Paulo]

What are your thoughts on culture-jamming?  Is it really effective?  Are corporations harmed by these jams?  Is it fair to target large corporations?  Does culture jamming do the opposite and actually benefit these corporations?  Is Adbusters too mainstream and expensive to be really considered a culture jamming entity?  Click here to read some criticisms of adbusters from Wikipedia.

Photoshop Assignment

Part 1
Critique of a culture jamming ad.
Find a culture jamming ad online, adbusters is a good place to start.  I also have some magazines at the front of the room from back issues of adbusters that I took out from the school library.  Go into your Google Classroom to work on the Critique.

Part 2

Success Criteria
– must be a parody of a specific organization advertisement (have the URL or magazine date referenced on your blog)
– if using photos that are not yours, make sure you share the link where you got it from, and the name of the photographer
–Photoshop skills need to be used well
– parody must be easy to understand and communicated well

[Link to Culture Jam Photoshop Rubric]

Exemplars
Kaitlin