Sunteți pe pagina 1din 10

Car Platooning

Investigating car pooling techniques in large Motorways Mondal, Mr. Joy Krishna jm12752@my.bris.ac.uk April 16, 2014

Contents

Abstract

Driving control in automobiles is an extensively studied problem in engineering. Its complexity arises from the fact that its almost impossible to predict with enough accuracy the behavior of other agents on an urban road. However such roads are only part of our transport infrastructure. There is a signicant amount of driving that happens in motorways that connect large cities. These roads are mostly empty and only occasionally have other vehicles in them. In other words the problem of control is much more constrained in motorways such as these, reducing the complexity of the problem. Also due to the trend in the primary sector of our economies which involve conducting activities in isolated regions far away from civilization, we have an increased requirement of delivering large amount of material over long distances. This means we need to eectively not only provide control for a single vehicle but for multiple of them traveling in a platoon. This study explores how we can eectively control such a platoon. We begin by rst taking into account all the physical constraints on a single car.We then try to nd the most stable velocity response for a car which becomes our velocity model for a single vehicle we use this to write a dierential equation for the acceleration. Since all cars besides the driver has the same dierential equation governing its behavior we take our model for single cars and scale it for n cars. We nally nd out how to express stable congurations for a platoon of cars and use that to assess the stability of our control algorithm.

Introduction

In 2013 a group in University of Oxford successfully managed to run a driverless car on a constrained legal road in the UK [?] [?].. Google has been working on a project called Google Driverless Car with an aim of creating autonomous vehicles [?]. Their work has provided such good results that it has caused the State of Nevada in US to pass laws allowing the use of driverless cars in urban road [?]. These are just two examples of various eorts worldwide to create autonomous cars on urban roads. The problem of making driverless cars are as old as cars itself and has been persistently studied by engineers over the last 50 years. Though the technology and area of research is very promising, autonomous cars on urban roads practically is something we are not going to see in the near future since their are many technical and legal challenges facing the adoption of widespread autonomous vehicles on urban roads [?]. Its hard to place responsibility with autonomous vehicle when things go wrong such as in case of an accident, and we cannot accuracy predict the behavior of pedestrian, bicyclist and other human drivers on an urban road. In such cases technology has not improved enough to surpass the ability of an human driver. With so many uncertain behavior to account for on an urban road we can see why even though we have being able to provide eective control for aircraft from as far back as 1930 [?], we still struggle to provide a good algorithm for automatic control for driving in cars even though cars are much more primitive and lack the sophistication of an aircraft. This begs the question - are there situations on roads when the level of uncertainty is minimal or reduced as with the case of aircraft ?. The short answer is yes, there are situations we can think of when driving on a road has fewer uncertainties associated with it. The most crucial assumptions on these types roads is that we do not have other actors on these roads besides ourselves. A practical example of such roads would be the M4 Motorway. These roads generally have the properties of connecting various large dense cities and usually follow a straight path between them. In 2011 about 25% of all distances traveled by automobiles in the United States were driven on such roads as the M4 Motorway [known in the US as the Interstate Highway System ][?]. China has built in the last decade about 85,000 kilometers of such road [?]. There are many factors contributing to rise of long distance travel via roadway. Improved fuel eciency in cars allows cars to travel more distance more cheaply, our need for resources such as fossil fuel and the depletion of them from easy to access places near to cities and industrial zones has caused us to create mining operations in more remote places. Globalization has lead to a economical reality where its easier to transport what you need from far away countries than be self-reliant. All of these tells us that this trend will continue in the near future as we create more roads to connect remotes regions on the planet. There are also entire independent network of roads which are built to transport goods inside large open pit mines, these roads have large number of vehicles traveling in groups from one point to another. These groups are normally called platoons and have became crucial for the mining industry.The cost of running such platoons is expensive, mainly because the vehicles used are not ordinary and require specialized drivers with training.This has 3

lead to interest in trying to automate the handling and driving of such platoons. The problem is in many cases the general methods used involve using a lot of fairly expensive equipments and building of special roads. In many cases these costs many became more than the cost of training individual drivers. This study tries to nd out an eective model that can be used to solve the general problem of driving platoons in empty roads to rst help the primary sector in automating their logistics and in that process hopefully reduce the percentage of drivers on empty roads. These eorts should provide more context and ease the transition for humanity to an eventual widespread adoption of driverless cars on urban roads.

3
3 .1

Assumptions
Road

The constraints of the road are important in determining how much complexity we are willingly to use for our: 1. We are not assuming the co-ecient of friction on the road is innite and their exists a acceleration on the road for which the no-slip condition will be violated. We will not be modeling cars that do slip however the no-slip condition will play an important role in determining the constraints of our model. 2. The Road is assumed to be a euclidean straight line. We will not be modeling roads that follow a curved path.

3 .2

Car

There are few key assumptions we have made in regards to the vehicle: 1. Presence of Rolling friction - There is energy losses due to hysteresis in the tyres of the vehicle and contact between the tyre and road due to electromagnetic potential. We assume such factors of energy loss are modeled using the co-ecient of rolling resistance/friction. 2. The Aerodynamic Drag due to air-resistance on the surface of the car will be modeled using the drag coecient. Since the car will not be traveling at relatively high velocities, this assumption should be accurate for the model. 3. For the leader car we are assuming that the driver will not exceed the legal speed limit on the road and the follower cars too will be modeled such that they do not exceed the speed limit 4. Modern Cars can have three types of transmission: (a) Automatic Transmission - the control input available to a driver in automatic cars are just brake and acceleration, clutch and gear control is done using whatever model that has been implemented by the manufacturer of the car and can vary however I think its reasonable to assume that the model adopted allows a high energy conversion between the engine and drive shaft. (b) Continuous Variable Transmission - which is implemented as an pulley with conical drives. (c) Manual Transmission- The driver is given full control of what gear could be used, and when to release or stick the clutch to the engine shaft. In Manual Transmission the number of gears states ( ratio) available is usually limited. For our model we are assuming the most common method of transmission in cars which is automatic even though there is a good chance that the many of the vehicles that might use our model have manual transmission since their usage is much more common in the primary sector.

Free Parameters and Symbol Denition

There are many free algebraic parameters that we are assuming will remain constant for our model. They are listed in Table 1. These parameters should not in theory aect the outcome of the model however it is advised to use the values that are most likely true in the physical world. The values we have placed are assumed to be temporary and are just shown as an example. The symbols mapping our parameters will be used to derive and write down all our dierential equations 1. Minimum Spacing between Cars when not moving - this determines what the distance between the cars is when the cars are not moving as time approaches innity. Note that the distance between the cars are dynamically changing at each time instant however their is a lower limit to how much the distance between the cars will be. In most cases it can be assumed to be zero however depending on your reference which could be a instrument on the car this parameter may need to be adjusted. For example if the the sensor for measuring the distance is in the center of the cars than surely the minimum distance between the cars should be greater than half the cars length. Even the situation of having cars bumper to bumper many not be desired and we many wish to have about a few meters of distance between the cars including the length of the car. 2. The Signal Range is the distance between the cars after which they cease to communicate. We are assuming that this value stays constant for our model. In the real world though, communication may break intermittently since propagation of a signal is aected by its initial ux luminosity which may not stay constant. The signal can also be reected, refracted, absorbed and scattered due to the medium it is traveling.We are assuming that such elements have an negligible eect on the parameter A. 3. Density of Fluid is the density of the uid in which the car is traveling in. This would usually be the density of air. The parameter will be used for modeling the aerodynamical drag. 4. Stead State Sensitivity is an important parameter. In may ways this is similar to the Integral tuning parameter in PID controller however the value for this should not be altered since changing it would result in weaker acceleration / deceleration response in our model. Name Coecient of Friction in Road Rolling Resistance Coecient Drag Coecient Density of Fluid Width of Car Height of Car Radius of Car Wheel Weight of each Car Maximum Horsepower Output of Engine Minimum Spacing between Cars when not moving Speed Limit Number of Cars in Platoon (Follower) Signal Range Steady State Sensitivity Value 0.7 0.01 0.29 1.2 2 2 0.5 37,280 50 1 27.7 1 100 0.05 Units nil nil nil Kgm3 m m m N W m ms1 Cars m nil Symbol Fr FRr FD D Cl Ch Cr Cw Cp Cm Rl N A B

Table 1: Table of Initial Parameters

Name Distance Speed Acceleration Distance of front car Speed of front Car

Units m ms1 ms2 m ms1

Symbol x x x u u

Table 2: Table of Dynamic Condition in Car Table ?? are the dynamical parameters that are present for each of the cars, the rst three parameters (x,x , x) determines the state of the car while the last three parameters (u,u )are the external input. These parameters are all time dependent and we can also write them as (x(t),x (t), x(t),u(t),u (t)). We can see just from the variables that the equation we will get will be a second order non-homogeneous equation of the form Equation (??). The equation will most likely be non-linear. Qx + Wx + Ex = P u + Lu Where: Q, W and L are all functions of several variables. In an ideal case the platoon cars will provide information about each others state. If we make the assumption that the platoon leader is the nth Car. The state variables of the nth car (xn ,xn , xn ) will became the input parameters for the n 1th car (un1 ,u n1 ). It does not matter if the ideal case is not followed. If the assumption is made that the cars are only tted with a device that measures distance to the car infront of it assumed h(t), we can nd out u by using equation (??) u(t) = x(t) + h(t) (2) (1)

We then use the backward dierence formula for nding derivatives to compute u shown in equation (??), assuming we are storing the value of h(t 1) and x(t 1) from the last time step. u (t) = Where: t is the time step or the reciprocal of the sampling rate of the device measuring h. x(t 1) and h(t 1) are the parameter values from the last time step. Regardless of how much information we are being given the underlying model is the same, with the only dierence being having to numerically nd some of the parameters. The reason we are not explicitly assuming equation (??) and (??) is because without them the model is much simpler to reason, express and analyze so that we can solve the more important problem of nding a dierential equation with stable properties for our acceleration control. [x(t) + h(t)] [x(t 1) + h(t 1)] t (3)

Measuring The State of the Platoon

We need to dene a variable that we can use to express/measure the state of the platoon, let us use a a platoon of three cars to illustrate this , and why we need this measure. Figure 1: Various states of a three vehicle platoon

State In Figure ?? We are assuming that all cars are traveling in the same velocity (v ) and d[n] is the distance between car [n] and car [n-1] in the platoon where n = 1 is the platoon leader. The car colored green is the platoon leader and the yellow cars represent the follower cars. From now on we will use the notation [S](d[n] ) to represent d[n] of State [S] and so forth( i.e A(d1 ) represents d1 for State A). dmin represents the minimum braking distance at the velocity the cars are moving. for our example we are considering A(d1 ) = A(d2 ) = dmin . The notion of space is expressed as area or volume and we can dene a boundary around our platoon Since we are considering the problem in 1 dimension we can dene this boundary as [S](d1 + d2 ) for each state S and can use the notation [S]D to express it, (i.e A(d1 + d2 ) = AD ). To help dene our measure we need to make a few observations: 7

AD > BD = CD = DD . Since AD is the minimum braking distance for both the cars, AD - BD > 0 by which we mean that AD - BD represents extra space. The importance of this extra space is an interesting question and it can be argued that this space provides a buer region for emergency braking , however we have already established AD as enough space required for the car to brake even in cases of emergency, keeping this extra space actually creates problem for the platoon in two major ways: 1. We are eectively increasing the probability of other cars from adjacent lanes to occupy this space hence further increasing [S ]D resulting in a divergence or space occupied by the platoon. 2. The wi/bluetooth network is limited and by increasing [S ]D we are increasing the chances of loosing communication among the platoon vehicles. By considering these assumptions we can intuitively say that State A is the most optimum state for the platoon to be in while State B, C and D represents sub-optimal states. We could use the measurement we are currently
n=2

using which is
n=1

dn to provide us with a measurement of state but this would not be eective since:

1. BD = CD = DD 2. State C and D are more similar than State A and B, in other words small changes when all the cars are near the platoon leader results in more state change than when they are further away, ( i.e having two of the platoon followers in our case at 100km away from the leader is at the same state as having them 50km near the platoon leader, while having the cars 10m from the leader is vastly dierent from having them 20m from the leader ). 5 .0.1 Square Measure
n=2

SD =
n=1

dn 2

(4)

This measure helps dierentiate between State A , B and C solving the rst problem ,however State B and State D are still equal ( we are assuming B(d2 ) = D(d1 ) and B(d1 ) = D(d2 )), we can overcome this problem by redening d. 5 .0.2 Relative Square Distance

Assume the maximum distance our platoon leader can see is wmax which is physically expressed as the distance at which communication breaks, in our model that would be A. We are going to dene a car platoon car as a car that is within wmax and wants to be part of the platoon which we have accepted to be part of the platoon. Withing this cars the car that is closest to wmax is also logically furthest from the platoon leader and hence this car would be our reference point or origin, we calculate the distance of each car from this car, square that value and then sum those values to give us the measure we need.
n=2

This measure solves the second problem we were having with using
n=1

dn and is a better version of the square

measure, its interesting to note the relative square measure is mathematically similar to statistical variance with mean of zero, however this is rather a deterministic model and even though the mathematical operation is the same the context is quite dierent. We can use equation (??) and see that the state of the platoon would be dened as the sum of the square of the solution of the vector dierential equation and subtracting from that that the minimum of the solution vector multipled by n.

Limitation

The major limitation of the model would be the lack of accountability for spare number of other cars on the road. We could also not use this model on a real car since in a real car the actual control we have is that of the change of the acceleration which is the derivate of the acceleration.

References
[1] http://mrg.robots.ox.ac.uk/robotcar/index.html [Date Accessed - 3rd April 2014] [2] http://www.bbc.co.uk/news/technology-21462360 [Date Accessed - 3rd April 2014] [3] http://www.ted.com/talks/sebastianthrungooglesdriverlesscar [Date Accessed - 3rd April 2014] [4] http://cyberlaw.stanford.edu/publications/automated-vehicles-are-probably-legal-united-states [Date Accessed - 3rd April 2014] [5] http://books.google.co.uk/books?id=4ykDAAAAMBAJ&pg=PA22&rediresc=y#v=onepage&q&f=false [Date Accessed - 3rd April 2014] [6] http://www.theatlantic.com/technology/archive/2013/10/the-ethics-of-autonomous-cars/280360/ [Date Accessed - 3rd April 2014] [7] http://www.fhwa.dot.gov/policyinformation/statistics/2011/vm1.cfm [Date Accessed - 3rd April 2014]

[8] http://nexus.umn.edu/Courses/ce5212/Case4/articles/Journal%20of%20Transport%20Geography,%20vol.%209%20(20 [Date Accessed - 3rd April 2014] [9] http://onlinepubs.trb.org/onlinepubs/sr/sr286.pdf [Date Accessed - 3rd April 2014]

List of Figures
1 2 Velocity Prole - Desired velocity R plotted against distance between the cars . . . . . . . . . . . . Various states of a three vehicle platoon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 13

List of Tables
1 2 Table of Initial Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Table of Dynamic Condition in Car . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 6

10

S-ar putea să vă placă și