Swarm Robotics in the Studio

J.Timmis1, E. Buchanan1, E. Clark1, H. Erdem1

B. Haworth2, A. Muhammad1, M. Mohammedshahid1, K. Nambiar1

M. Parmar1, J. Hilder1 and T. Sharpe2

Abstract— In the creative industries, understanding the fu-ture demands and desires of people allows you to stay onestep ahead. We are interested in understanding what affect thepresence of robotic devices will have on the day to day activityof people, in a work environment, so that we might understandhow technology could be best incorporated into peoples lives. Todo this, we have developed a novel, interactive robotic swarm,that is able to sense various features of a working environmentand respond accordingly. The system is simple to install, lowcost and is easy to operate. Robots in the swarm move fasterwhen people are moving, or perform various light displays whenthe noise level increases, and even tweet its own state whenthings are changing rapidly in the studio. We believe this willbe of the first examples of a deployed, interactive swarm in astudio environment. We will be undertaking longitudinal studiesto assess the affects of introducing the swarm and how this haschanged perception of technology and affected productivity.

I. INTRODUCTION

Work in this paper presents a collaborative effort betweena creative agency, The Beautiful Meme, and robotics studentsand researchers at the University of York. Our objective is tobuild and install a swarm robotic system that will reside inthe studio of The Beautiful Meme that is responsive to theenvironment and alter its behaviour based on a number ofdifferent triggers from the environment. The project serves anumber of purposes, from providing the basis to understandhow the deployment of such technology might help peoplefrom the creative industries understand what it is like to havedynamic, interactive and mobile technology present in theirworkplace, and provide a focus for students learning aboutrobotics.

There are a limited number of examples of interactiveswarms, most of which focus on robots being sociallyinteractive, for example [2] where focus tends to be ondeveloping socially able robots that are capable of interact-ing in a socially acceptable manner, with themselves andhumans. However, our work does not focus on this aspectof socially interactive swarms, we are more interested in theconstruction of a simple, low cost platform for exploringsimple interactions in a daily work environment, whilst at thesame time providing a usable platform for teaching purposes.

Our specific questions relating to this work are: 1) whatare the requirements for a responsive and interactive swarmand 2) what is the minimal processing required to facilitateinteraction from motion and sound with the swarm and 3) can

1York Robotics Laboratory, University of York, UK.jon.timmis@york.ac.uk

2The Beautiful Meme, UK.

users interact with the swarm to directly affect behaviour. Inthe longer term, we will address the issues of acceptabilityof the robotic swarm in the workplace, but that is not thefocus of this abstract. In order to address our questions, wehave developed an interactive robotic swarm using the Pi-swarm robot [1] developed at York as the platform. Thisrobot is sufficient for our needs at present, and serves as aprototype system for interaction. It may be in the future, weexplore different platforms for interaction, but that wouldbe the subject of further work. We selected the Pi-swarmdue to the fact that it is mobile, simple to programme, issmall (so deployable on a table in the studio environment,a key requirement) and is extensively used by the studentsin their studies, so there is a degree of familiarity withthe platform. The swarm is deployed on a table of size140cm x 80cm that has attached a variety of sensors tomonitor the environment which in turn affect the swarmbehaviour. We believe this represents the first interactiverobotic swarm installation. The rest of this extended abstractis structured as follows: section II details the requirementsfrom the users perspective and how these are translated tosystem requirements, section III outlines the system designincluding the hardware, infrastructure and algorithms andsection V draws some initial conclusions from the work.

II. REQUIREMENTS FOR THE SWARM

A number of functional requirements were established forthis work, specifically:

• The robot swarm should be able to flock and/or maintainmovement during certain periods. It might be the casethat robots can be restful, and the robots adapt theirmovement amount by activity surrounding the table

• The robot swarm should respond to movement in theenvironment: The swarm should start at rest and whenmovement in the studio increases, as monitored bymotion sensors, then the speed of the swarm shouldincrease

• The robot swarm should be able to respond to soundlevels in the environment: Robots in the swarm areequipped with a range of LEDs which will be used invarious ways to represent a response to the noise levels

• The robot swarm should know when to enter a sleepmode. The swarm should enter a sleep mode wheneither it is the end of a working day, or there has beenno movement in the environment for a predeterminedamount of time

• The robot swarm should be able to access social media,specifically, Twitter: The state of the swarm should berecorded by a central PC and periodically, or when achange in state is detected, a tweet is sent out from apredefined list of possible tweets

III. SYSTEM DESIGN

Pi-Swarm robots, as seen in Figure 1, are the hardwareplatform used for this study. They comprise a Pololu 3-Pi robotic base with an MBED-based expansion board.The robots are a low-cost yet versatile design, specificallycreated for swarm interactions, allowing simple RF-basedcommunication between robots and the outside world andproviding a number of environmental sensors and actuators toallow interactions with people in the studio [1]. A controllerboard establishes a link between the swarm and additionalmotion and sound sensors, and an external computer providesaccess to the internet allowing social media interactions fromthe swarm, specifically Twitter.

Fig. 1. The Pi Swarm robotic platform

The swarm is placed on a table which has a variety ofmotion sensors, microphones and infra-red beacons attached,see figure 2 for image of the infra-red beacons attachedto the table. The infrared beacon along one edge of thetable emits 50ms long pulses once every second. The Pi-swarms can differentiate between the signal from the beaconand other Pi-swarm robots in the swarm. The beacon signalis used to synchronise the swarm and calculate orientationfor each robot. Passive infra-red (PIR) motion sensors andmicrophones are placed on the sides of the table to monitorfor movement of people, and noise, around the table. ThePIR sensors produce a logic high output when motion isdetected. The microphones are fed into a peak-holding am-plifier circuit, which is bandpass filtered to respond to voicefrequency signals between approximately 300Hz and 3kHz.Data from these sensors is sent to a central microcontrollerwhich process the data to analyse changes in motion overtime, see figure 3. The results of which are transmitted

using RF to the robots and used to alter parameter settings,which in turn will affect their behaviour, with robots movingtowards the location of movement.

The default operation of the swarm is to flock. This isimplemented via a simple boids algorithm, with the param-eters of the speed and orientation being affected by valuesbeing recorded from the environment. Should a person movein front of the PIR then the trajectory of the robots in theflock are altered, incrementally, to head towards the PIRsensor area. In addition to that change, local effects alterthe behaviour of other robots through the normal flockinginteraction rules. Should noise occur in the area, then thiswill effect not only the speed and coherence of the robot, andhence other robots indirectly (again via the interaction rules),but they also initiate a simple LED light display in response.Finally, over time, as the state of the swarm changes, forexample, from moving slowly in the centre of the tableto moving towards a PIR sensor, the swarm will tweet thechange in status via its own Twitter account. Tweets to theTwitter feed are made on an ad-hoc basis.

Fig. 2. The robots synchronise and calculate their bearing by detectingpulses from an IR beacon along one edge of the table.

Fig. 3. The PIR sensors (left) and microphones are attached to amicrocontroller board (right) that sends messages to the swarm .

A. Behaviour

We adopt a finite state machine approach for the imple-mentation of a simple behaviour-based architecture. Figure

4 shows two main states of operation: Aggregation andFlocking. Switching occurs between these two states basedon the sound level that is received by the robots from thetable controller, as outlined above. Therefore, if the soundlevel is high (which depicts a noisy environment) the robotswill be in the flocking state and move around the table,whereas when sound level is low (quiet environment) , therobots will switch into aggregation state and congregatetogether. The speed of operation during both states is affectedby the amount of motion detected around the table, asoutlined above.

Flocking) Aggrega.on)Sound)<)Threshold)

Cohere)

Separate)

Cohere)

Avoid)

If)distance)<)d_threshold)&))Sound)>)s_)threshold)

If)distance)<)d_threshold)&))Sound)>)s_)threshold)

Sound)>)Threshold)

Fig. 4. Basic state machine for individual robotic units

B. Reacting to sound

Sound is captured by an Arduino board using the soundsensor. A sample of sound is collected for 50ms and thedifference of the highest and lowest value of the sampleis calculated. This value is then subjected to a thresholdvalue which is determined through parameterisation basedon the background noise levels in the environment. If thepeak to peak value crosses the threshold, a HIGH value istransmitted to the control board by simply making a GPIOpin of Arduino HIGH, otherwise LOW. The control boardreceives signals from the Arduino every 1 second. Uponreceiving a HIGH signal, it increases the level variable ofsound by a step of 0.1until it reaches a maximum of 1.0,which means the environment has turned noisy. Similarly,if control board receives a LOW signal, it reduces the levelvariable by 0.07 until it reaches the minimum value, 0. Theslow decay and improvement results in a gradual changeof behaviour and add persistence to the swarm reaction tosound. It is important to note that the robots change theirbehaviour only when the value reaches either of the terminalvalues, that is either 0 or 1. This value is transmitted to therobots using RF communication.

C. Reacting to motion

As previously stated, the table incorporating the robots has3 motion sensors on three sides. Readings are taken fromthese sensors every 200ms. Every motion sensor has a levelvariable attached to it which varies from 0 to 10. For everyHIGH on a motion sensor, the corresponding level variableis incremented by 1 and for every LOW, it is decrementedby 1. All the levels are then added and normalised to get the

Fig. 5. Example run from robotic swarm reacting to motion around thetable

speed parameter between 0.05 and 0.20. This value is sent tothe robots via which then update their speed proportionally.

IV. RESULTS

By means of example, figure 5 shows recorded data fromthe operational robotic swarm, and videos from example runscan be found on our website1. The data shows that, over time,the speed of the robotic swarm changes over time, increasingand decreasing with respect to motion sensor data recordedfrom around the table. Similar results are also obtained fromthe sound sensor data, with respect to aggregation, as can beseen in the videos.

V. CONCLUSIONS

We have developed a simple, low cost robotic swarmwhich we feel is a useful platform for the investigationinteractive robotic swarms. The simple interface to the pi-swarm makes it easy to programme and flexible, along witha single table infrastructure, rich behaviour from the swarmcan be observed. The next stage of the work is to deployin a studio environment, and instigate a long-term studyon the effects of locating such a robotic swarm in a placeof work. Staff working in the studio will be subject to alongevity study, initially ascertaining expectations from theswarm, and requested to keep a log book of interaction overtime. Activity data will be recorded from the swarm andmonitored on a weekly basis and analysed. A final interviewwill be taken after an agreed period of time to follow-up onthe initial interview.

ACKNOWLEDGMENT

This work is funded by The Beautiful Meme, York, UK.Jon Timmis is part funded by the Royal Society.

REFERENCES

[1] J.A. Hilder, R. Naylor, A. Rizihs, D. Franks and J. Timmis, ThePi Swarm: A low-cost platform for swarm robotics research andeducation, Accepted for the 15th Towards Autonomous RoboticSystems Conference TAROS 2014.

[2] T. Fong, I. Nourbakhsh, K. Dautenhahn. A survey of socially interac-tive robots Robotics and Autonomous Systems, 42(34):143-166 (2003)

1https://www.york.ac.uk/robot-lab/interactive-swarm/