Wireless Network for Mining Applications

Gerold Kloos, Jose Guivant, Stewart Worrall, Andrew Maclean and Eduardo Nebot

ARC Centre of Excellence for Autonomous Systems

School of Aerospace, Mechanical and Mechatronic Engineering

University of Sydney

NSW Australia

{g.kloos, jguivant, s.worrall, aj.maclean, nebot}@acfr.usyd.edu.au

Abstract

This paper addresses the problem of people -machine interaction presenting a framework totrack the positions of interacting local agents ina mining type environment. The approach pre-sented allows for the incorporation of additionalsensing technology to add integrity and relia-bility to the tracking process and potentiallyreduce the uncertainty in the estimation of po-sition of the agents. The system is based on aprinciple that we call virtual networks. Thesenetworks are formed in a local area where theagents interact. Each agent broadcasts a seriesof messages and updates the position informa-tion of the surrounding agents using statisticalfiltering techniques and models that best pre-dict their behavior. The concept can also beapplied to move information using agents thatoperate in larger domains. Experimental re-sults of people tracking in mining applicationwill be presented to demonstrate the main con-tributions of this paper.

1 Introduction

There is a wide variety of applications for systems thatcan track the position of agents, like persons or vehi-cles in a reliable and predictable manner. Most indoorapplications focus on improving the interaction of mo-bile robots with human beings therefore requiring goodknowledge of persons locations. Outdoor applicationsof position tracking are more common in the areas ofsafety and surveillance. Although the results presentedare applicable to a large variety of processes we focusthis work in outdoor mining type environments. One ofthe main problems in open pit mining is the machine-machine and machine-people interaction. Mining pro-cesses require the movement of ore from the pit to thecrusher. This process involves truck loading, that is in-teraction of shovels / front end loaders with trucks, Fig-

Figure 1: Interaction between large mining machines.

ure 1, and interaction of haul trucks with vehicles andpeople, Figure 2.

The relative position and orientation of large machinesis usually desired to improve productivity. For example,by achieving the right orientation and position of a trucknext to a shovel the loading time could be reduced thusmaximizing capital utilisation. The interaction of largemachines and utility vehicles and people is usually im-portant for safety reasons. Drivers of large off-road haultrucks often cannot see objects (personnel, utility vehi-cles, etc) in close proximity to the truck. The zone di-rectly in front of the truck and adjacent to the non-driverside of the truck poses the greatest risk for an accident.These blind zones have been the cause of several haultruck/utility vehicle accidents and near misses. Of par-ticular concern is the ability of truck drivers to verifythat these zones are clear before pulling away from astationary position. Several technologies had been pro-posed to solve this problem:

1. GPS only based systems - this approach requires allmobile equipment and personnel to be fitted withGPS units that communicate with a base station.

1

Figure 2: Interaction between Haul Trucks and People.

This solution is expensive and not failsafe since itrelies on each object having an operational GPS unitand complete GPS coverage in the pit. AdditionallyGPS systems will in many cases have a poor satelliteavailability when the operator is very close to largemachines.

2. RF-ID tags - This requires all mobile equipmentand personnel to wear a RF-ID tag that will re-spond when in the truck proximity zone. Althoughthere are a few commercial systems based on thisapproach, most of them only provide an indicationthat an agent is in the area. The detection of thelocation of the agent requires additional hardwareand accuracy and cost may be an issue.

3. Video systems - this requires computer based im-age recognition of an object in the proximity zone(which can be unreliable in varying light condi-tions). Alternatively, this approach requires thedriver to see the intrusive object in a small cab-mounted video screen that in poor lighting condi-tions can be unreliable.

4. Radar / camera / lasers based systems. These ap-proaches are based on range and bearing sensors ac-tively detecting an object in proximity to the trucks.One promising approach [Ruff, 2004] utilises a lowfrequency short range radar. A commercial versionof this type of radars operates at 5.8 GHz and candetect objects at ranges up to 8 meters within andarc of 55 degree horizontal and 20 degrees vertical[Preco Electronics, 2001]. It is fitted with hardwareto provide visual and audio alarm if an object iswithin the area. This output is then used to turnon the appropriate camera. Since the resolution ofthe radar is very poor at this frequency it is verydifficult to avoid false alarms. The resolution could

be improved by using MMWR or lasers. MMWRare still very expensive and lasers are not an optionin this type of applications due to the environmentalconditions.

There are a large numbers of publications that addressthe problem of people tracking in indoor environments.These approaches are often vision-based, like in [Chooand Fleet, 2001] and [Poon and Fleet, 2002], which alsorely on modelling the human’s body. The co-operationof heterogenous sensors like vision and laser to track per-sons in office environments is also presented in [Brooksand Williams, 2003]. [Schulz et al., 2003] and [Hahnelet al., 2004] use RF-ID technology to localize personsin office environments. [Liao et al., 2004] apply RAO-Blackwellised particle filters to raw GPS data to learnand infer persons travelling modalities and route. Aninteresting outdoor application is described in [Juang et

al., 1996], where the authors present algorithms to trackanimals positions’ in very wide areas. [Patterson et al.,2003] presents an approach where a model of the trav-eller moving in an urban environment is learned fromGPS data.

[Nieto and Dagdelen, 2004] developed a GPS andIEEE802.11 wireless technology based system to trackthe position of trucks and vehicles in open cut mines.They also provide the truck driver with real-time graphi-cal information about his relative location to other trucksand its own location is projected on a 3D map of themine. This approach is based on the availability of highaccuracy GPS information.

Most of the approaches presented before addressedvarious aspects of the people tracking problem and in-cluded in most cases experimental results to validate thealgorithms or technology in constrained environments orin applications where reliability and integrity is not animportant issue. A different case is the proximity prob-lem in mining applications [Ruff, 2004]. In this case weneed to guarantee that algorithms and sensors will pro-vide an acceptable degree of uncertainty while operatingunder all possible conditions. By looking at the con-straints of the problem it is clear to infer that any solu-tion based on a single sensor will fail at a certain point orwill not provide enough information to detect all possiblethreats. We believe that a successful implementation of aproximity system will consist of multiple sensors workingtogether to achieve maximum protection for the people-machine interaction task. This is in contrast to most ofthe already available proximity systems which are basedon a single sensor. The use of sensor-information comingfrom multiple sensors will complement the capabilities ofeach sensor to create a more reliable tracking system.

The paper is organised as follows: Section 2 describesthe basic concept of virtual network and the actual sys-tem implementation. Section 3 presents the model used

2

to track agents and experimental results of people track-ing using the concept with a single sensor are shown insection 4. Finally in section 5 conclusions are drawn andfuture work is presented.

2 Virtual Network System

The approach proposed is based on the concept of amine wide virtual wireless system currently under de-velopment in a CRC Mining project. The fundamentalconcept is based on the creation of ad-hoc temporary dy-namic networks. Under the IEEE’s proposed standardfor wireless LANs (IEEE 802.11), there are two differentnetwork configurations: ad-hoc and infrastructure. Inthe ad-hoc network, computers are brought together toform a network in real time. Figure 3 shows the struc-ture of this network where each element is potentially incontact with any other element within the range in thenetwork.

M3

802.11b

M2 M5

M1

M4

Figure 3: Ad-Hoc Network: A temporary network isformed in a local area.

IEEE 802.11x technology forms the core of our sys-tem. Each agent in a local ad-hoc network builds upan information table with all current information aboutitself and other agents. At regular intervals the nodesbroadcast its information and make it available to itsneighbors (see Figure 3). The information transfer be-tween different local ad-hoc networks which are not per-manently interconnected will be achieved through fastmobile agents (Haul Trucks), that transport the infor-mation from one local ad-hoc network to another (seeFigure 4). In doing so it opens up the possibility to

transport all kind of information, e.g. the positions ofagents, to all members of the whole network. With somelatency all members in the network will have all informa-tion available for them. With this approach each agentcan be retrofitted with a variety of sensors and absoluteand relative information can be transmitted to the net-work. Further decentralized processing can be appliedto obtain the most likely position of each agent based onall the information available in the network.

The system is very scalable since it does not requireany infrastructure node to operate. It allows for theimplementation of multiple-sensor-based proximity sys-tem and automated data retrieval from remote areas (atruck passing by can collect data from a remote sen-sor). The appropriate integration of these ideas has thepotential to become the main platform for moving infor-mation around the workplace without having to installa full wireless coverage. This is particularly attractivefor mining application where large areas of operationare required but the activities are concentrated in fewlocations.

Truck

802.11b

Pers

GPS

Pers

GPS

Mobile

GPS

Pers

GPS

Pers

GPS

Figure 5: Ad-Hoc Network. Although two people agents(dark background) are not directly in contact with thetruck, their position is also available through the otheragents in the local network.

2.1 System Implementation

GPS and Wireless capabilities are becoming common inmany system components. These capabilities are alsoemerging as standard options on many pieces of miningequipment. There are now a number of new processors

3

network backbone

Base Station

Shovel

Legend:

Truck / light vehicle

pedestrian (with GPS)

local network zone

haul road

802.11 connection

Figure 4: Proximity detection and information transport.

which implement standard IEEE 802.11x wireless com-munication which can be easily integrated with GPS andother sensors. These can be attached to compact andruggedised devices on personnel and mobile equipmentto be used in harsh environments like mining. Thesecomponents allow individual nodes to provide informa-tion to other in-range nearby nodes and more impor-tantly have the potential of building a wide area networkof the type described above.

This implementation of the virtual network conceptuses GPS receivers to obtain the location of agents (seeFigure 5). Each agents processes all the informationavailable in direct or indirect form, determines its mostlikely position and broadcasts it at regularly intervalsthrough the network. Additional sensors like the onesdescribed in section 1 can easily be added to this sys-tem and their information can also be shared with otheragents in the network.

The hardware used consists of PDA’s for the peopleagents and PC/104 based systems for the other mobileagents. All agents are equipped with GPS receivers. Ta-ble 1 presents the hardware used for each type of agent.

Table 1: Overview of agent-types and hardwareAgent type Computing platform Sensor

People PDA with Bluetoothand IEEE802.11

GPSReceiver

Truck PC/104 withIEEE802.11

GPSReceiver

Lightvehicle

PC/104 withIEEE802.11

GPSReceiver

Figure 6 shows a people agent with a PDA and a GPSunit.

Figure 6: Agent of type people with PDA and GPS. TheGPS unit is attached to the helmet.

Other agents have operator interfaces according to theapplication.

3 Tracking models

In this type of applications it is usually possible to im-prove the quality of the estimation by using appropriatemodels to predict the position of agents. Furthermore insome cases the sensory information will not be availableand the prediction models will be the only source of in-formation. There is a significant number of publicationsthat address modelling of trucks and vehicles. Proba-bly the simplest model is the one that assumes that themobile agent is moving at constant velocity. Modellingpeoples’ behaviour is much more difficult and requiresmore elaborated models and estimation algorithms. Inthis work we use the direct discrete constant accelerationmodel with a Kalman filter [Bar-Shalom et al., 2001] to

4

predict and estimate the position of the people agents.We use a decoupled model for the x and y coordinate.

The state prediction equation is

x(k + 1) = Fx(k) + Γv(k) (1)

and we define the state vector to be

x = [x, x, x, y, y, y]T

(2)

where the state transition matrix F is given by

F =

1 T 1

2T 2 0 0 0

0 1 T 0 0 00 0 1 0 0 00 0 0 1 T 1

2T 2

0 0 0 0 1 T

0 0 0 0 0 1

(3)

and the noise Γ gain by

Γ =

1

2T 2 0T 01 00 1

2T 2

0 T

0 1

(4)

The noise covariance σ2v

multiplied by the gain Γ isgiven by:

Q =

1

4T 4 1

2T 3 1

2T 2 0 0 0

1

2T 3 T 2 T 0 0 0

1

2T 2 T 1 0 0 00 0 0 1

4T 4 1

2T 3 1

2T 2

0 0 0 1

2T 3 T 2 T

0 0 0 1

2T 2 T 1

σ2

v(5)

Note that T denotes the sampling period.This model will not be appropriate in cases where the

operator is working close to the truck since in this casemany manuevers will not be able to be characterizedby constant velocity models. Actual research is underprogress to recognize these situations and apply othermultiple hypothesis algorithms such as particle filters.

4 Experimental results

An outdoor experimental data set was obtained using aniPAQ-client retrofitted with an EMTAC BT-GPS. Thelog-file was about 15 minutes in duration and logged un-der good GPS conditions, i.e. good sky visibility. TheGPS was working in autonomous mode and its informa-tion was logged at a 1 second sampling time interval.The model used to track the agent in the Kalman filterwas the direct discrete constant acceleration model as

State−Estimate (Position) and GPS−Data (Position)

x−position (m)

y−po

sitio

n (m

)

−20 −10 0 10 20 30 40

−40

−30

−20

−10

0

10

20

Figure 7: Trajectory of the agent: Red is GPS truth,blue is estimated path.

State−Estimate (Position) and GPS−Data (Position)

x−position (m)

y−po

sitio

n (m

)

32 34 36 38 40 42 44 46−16

−14

−12

−10

−8

−6

−4

−2

0

Figure 8: Zoom of the track. Red is GPS truth, blue isthe estimated path.

described in section 3. Some preliminary results are pre-sented in Figure 7, where GPS and estimated positionare shown in red and blue respectively.

As expected, the maximum errors are present in thecorners (see Figure 8). This is due to the fact that a con-stant acceleration model will be very inaccurate in thiscase. As mentioned before these situations need to bedetected and different filtering schemes need to be acti-vated to at the very least be able to obtain conservativeestimates of the agents’ uncertainty.

The covariances are depicted in Figure 9. By inspect-ing the innovation sequence one can see that the differ-ence between prediction and observation is never big-ger than 3 m (see Figure 10). This graph also showsthat about 95% of the innovations fall within a 2-σ gate,

5

Covariances (x−pos/y−pos)

time (s)1.335 1.335 1.335 1.335 1.335 1.335 1.335 1.335 1.335

x 1011

0

0.2

0.4

0.6

0.8

1

1.2

1.4

Figure 9: Covariances - note that both x-covariance andy-covariance are identical.

Innovation (y−pos/y−pos) and Innovation stddev (x−pos/y−pos)

time (s)1.335 1.335 1.335 1.335 1.335 1.335 1.335 1.335 1.335

x 1011

−4

−2

0

2

4

6

Figure 10: Innovations: red x-coordinate, blue y-coordinate.

which is an indication for the consistency of the filter.

5 Conclusions and Future work

This paper presented the implementation of a frameworkto enable the solution of the proximity problem in a re-liable manner. Experimental results were presented tovalidate the concept of the virtual network with an ex-ample using a single sensor, in this case GPS. Researchin progress is addressing the integration of networks ofclose proximity programmable range only sensors. Thefull integration of the tracking system will allow to im-plement sensor management techniques to exploit thesensor’s capabilities in an optimal manner according tothe actual threat situation.

Acknowledgments

This work is supported by the CRC Mining, the Aus-tralian Research Council (ARC) Centre of Excellenceprogram and the New South Wales Government. Alsospecial thanks to QSSL for donating copies of the QNXMomentics Professional development system used to im-plement the real-time data acquisition system for theexperiment.

References

[Bar-Shalom et al., 2001] Yaakov Bar-Shalom, X.-RongLi, and Thiagalingam Kirubarajan. Estimation with

Applications to Tracking and Navigation. Wiley-Interscience, New York, N.Y., 2001.

[Brooks and Williams, 2003] Alex Brooks and StefanWilliams. Tracking People with Networks of Heteroge-

neous Sensors. Australasian Conference on Roboticsand Automation (ACRA), Brisbane, 2003

[Choo and Fleet, 2001] Kiam Choo and David J. Fleet.People Tracking Using Hybrid Monte Carlo Filtering.IEEE International Conference on Computer Vision,Vancouver, 2001

[Hahnel et al., 2004] Dirk Hahnel, Wolfram Burgard,Dieter Fox, Ken Fishkin and Matthai Philipose. Map-

ping and Localization with RFID Technology. Interna-tional Conference on Robotics & Automation (ICRA),New Orleans, 2004.

[Juang et al., 1996] Philo Juang, Hidekazu Oki, YongWang, Margaret Matonosi, Li-Shiuan Peh and DanielRubenstein. Energy-Efficient Computing for Wildlife

Tracking: Design Tradeoffs and Early Experiences

with ZebraNet. International Conference on Architec-tural Support for Programming Languages and Oper-ating Systems (ASPLOS), San Jose, 2002.

[Liao et al., 2004] Lin Liao, Dieter Fox and HenryKautz. Learning and Inferring Transportation Rou-

tines. National Conference on Artificial Intelligence(AAAI), San Jose, 2004.

[Nieto and Dagdelen, 2004] Antonio Nieto and KadriDagdelen. Development and Testing of a Vehicle Col-

lision avoidance System Based on GPS and Wireless

Networks for Open-Pit Mines. International Sympo-sium on Computer Applications in the Minerals In-dustries (APCOM), South Africa, 2003.

[Patterson et al., 2003] Donald J. Patterson, Lin Liao,Dieter Fox and Henry Kautz. Inferring High-Level-

Behaviour from Low-Level Sensors. InternationalConference on Ubiquitous Computing (UBICOMP),Seattle, 2003

[Poon and Fleet, 2002] Eunice Poon and David J. Fleet.Hybrid Monte Carlo Filtering: Edge-Based People

6

Tracking. IEEE Workshop on Motion and Video Com-puting, Orlando, 2002.

[Preco Electronics, 2001] www.preco.com.

[Ruff, 2004] Todd M. Ruff. Recommendation for Test-

ing Radar-based Collision Warning Systems on Heavy

Equipment. Annual Institute on Mining Health, Safetyand Research, Salt Lake City, 2002

[Ruff, 2004] Todd M. Ruff. Advances in Proximity De-

tection Technologies for Surface Mining. Annual In-stitute on Mining Health, Safety and Research, SaltLake City, 2004

[Schulz et al., 2003] Dieter Schulz, Dieter Fox, JeffreyHightower. People Tracking with Anonymous and ID-

Sensors Using Rao-Blackwellised Particle Filters. In-ternational Joint Conference on Artificial Intelligence(IJCAI), Acapulco, 2003.

7