Motivated by anecdotal evidence, we hypothesize that an egocentric approach is more appropriate and relevant to providing fuel efficiency feedback than a systemic approach. In this paper we describe a proposed study to test this hypothesis, and present the design of a fuel efficiency feedback system for public transit bus drivers.
In response, operations attempted to “calibrate” the system by tweaking its thresholds. The result was that the feedback became useless and largely inaccurate, ultimately resulting in the abolishment of the system. In our discussions with the transport authority, it became clear that in addition to the misinterpretation of the feedback by the professional drivers as a rating of their driving, the mountainous terrain of Madeira caused genuinely inefficient driving. There was simply no way to avoid steep hills that took a significant toll on fuel consumption, thereby skewing the feedback towards inefficient driving. The attempts at calibrating the sysCopyright c 2011 for the individual papers by the papers’ authors. Copying permitted only f or private and academic purposes. This volume is published and copyrighted by the editors of PINC2011. tem failed because, effectively, the on-board equipment measured pure fuel consumption which in turn was intricately related to the steep terrain of the environment. On the other hand, drivers perceived the feedback as a reflection of their skills.
Our anecdotal experience with the public transport authority’s feedback system caused us to hypothesize that providing feedback on specific driver behaviour, as opposed to overall fuel efficiency, may be a more appropriate way for motivating driver behaviour change. Adopting a systemic approach to this issue, we argue that existing feedback mechanisms relating to efficiency provide a view of the complete system, parts of which the driver has simply no way of effecting (such as the steep terrain). Hence we argue that efficiency feedback focusing on parts of the system that the driver can effect (such as acceleration) may result in more efficient driving behaviour. We term this approach to feedback egocentric.
In this paper we describe a fuel efficiency reporting
and advisory system that takes advantage of the
multisensor and interactive nature of modern smart-phones
to present feedback to drivers. More specifically, we
are interested in deploying the system in public transit
buses to measure its effectiveness on positively
influencing drivers’ behavior. By continuously capturing
realtime sensor data, we can calculate the Vehicle Specific
Power (VSP), a surrogate variable that strongly
correlates with both fuel consumption and pollutant emission
levels, providing a systemic view of efficiency [
Research suggests that it is possible to achieve up to
15% of fuel consumption decrease when appropriate
driving behavior is used [
Raw fuel consumption levels and pollutant emissions
can be calculated through the use of Portable
Emissions Measurement Systems (PEMS). These are
connected to vehicles through their On-Board Diagnostic
(OBD) interface, letting the PEMS system access the
vehicle’s on-board computer and calculate multiple
parameters [
The Vehicle Specific Power (VSP) approach is used to
approximate and predict actual emissions levels and fuel
consumptions [
V SP = v ∗ (a + g ∗ sin(ϕ) + rcoef ) + acoef ∗ v3 (1)
where v is speed in m/s, a is acceleration in m/s2, g is
9.807 m/s2, ϕ is the road gradient value, rcoef is the
rolling resistance term coefficient, and acoef is the air
drag term coefficient. Another characteristic of VSP
is its ability to support payload modeling, especially
important in situations where this value has noticeable
impact, such as is the case with public transit buses
[
Devices such as smart-phones possess a wide variety of sensors, like GPS and accelerometers, that enable calculation of vehicle dynamics and consequently VSP values. It is then possible to approximate fuel consumption using solely internal smart-phone sensors. These devices can be easily incorporated into vehicles, and their ability to provide a rich and extensible interaction platform make them a feasible alternative mechanism to provide drivers with fuel efficiency feedback. Furthermore, and comparing with usual commercial systems such as Scania Fuel-Saving Driver Support System1, smart-phones are not restricted to specific vehicles, and can even be device independent, which is the case when using development platforms such as Google’s Android.
Receiving timely feedback is key to motivating behaviour
change, people need to be aware of their behaviour in
order to change it. Fischer found the most successful
feedback was given frequently, clearly presented, used
computerised tools and allowed historic or normative
comparisons [
Consolvo, McDonald, and Landay [
First, we make use of abstractions rather than counting
solely on raw data to display to drivers. Secondly, the
data shown should be unobtrusive. This is of paramount
importance for safety reasons, as we need the mobile
display to support ignorability and not distract the driver
unnecessarily. Thirdly, since the data is to be presented
in public, we need to present it in a way that the driver
will not feel uncomfortable if others are aware of it.
Fourthly, we decided to ensure that only positive
feedback is given, not punishing any “bad” behavior.
Concretely, we aim at rewarding possible low consumption
levels, but not use punishment for poor performance.
This decision is supported by the notion that positive
feedback can indeed increase intrisic motivation by
affirming competence [
We propose an experimental approach to study to what extent we can, through the use of egocentric feedback, 1http://www.scania.com/media/featurestories/sustainability/every-drop-of-fuel-counts.aspx influence public transit bus drivers driving behavior. In our study we are interested in the following research questions: • Can we accurately establish driving behavior profiles for bus drivers through the use of VSP calculations? • To which extent can we positively influence driving behavior through the use of egocentric feedback techniques? • Is the use of real-time more effective than the use of historic feedback, or is a combination of the two approaches most effective? Consequently, and based on the previous mentioned research questions, we raised the following hypotheses: • H1. The use of the VSP surrogate variable (and its derivatives) allows for accurate driving profile characterisation • H2. The use of egocentric driver feedback improves average fuel consumption levels • H3. The use of real-time feedback does not significantly influence driving behavior To test these hypotheses we propose to develop an Android based software to continuously collect sensor information so that trip instantaneous parameters, such as speed and acceleration, can be calculated. We will also consider the use of additional variable(s) to model the influence of passenger payload on the overall vehicle weight. Then, we intend to install equipment on-board public transit buses and calibrate the VSP model. The ground truth establishment of instantaneous fuel consumption levels is a necessary condition for the success of the VSP model. This may be achieved through the use of a PEMS system or a similar mechanism. Subsequently, we will develop a derivative of VSP called egoVSP, which ignores road gradient and is defined as follows egoV SP = v ∗ (a + rcoef ) + acoef ∗ v3 (2) Terms of the equation are defined equally as in eq. 1. These two fuel efficiency models, VSP and egoVSP, are one of the two variables we intend to manipulate in our study. The other variable is the type of feedback to provide: real-time versus historical. Table 1 shows the possible combinations of these two variables.
As it stands, the system is a working prototype. Targeted mainly at public transit bus drivers, the system RawInput Sensor Data
Real-time Processing
Pipeline Comp1onent . .
CompNonent
Transformed Output Storage
User Interface Real-time feedback
VSP Acc. 10l 50 0.5 Historical feedback
Driver is flexible and extensible enough to provide support for any kind of vehicle.
An overview of the architectural design is seen in Fig. 1, where the mechanism that is used to produce the final output to the driver is visible. Raw sensor data is sampled at several times per second, before it is passed to a real-time processing pipeline. This allows us to execute tasks in parallel that may require some computational complexity, therefore increasing system overall speed and responsiveness. The advantage of such a scheme becomes more evident when, for example, the system is required to perform continuous sensor data integration by means of a Kalman Filter.
The calculation of the vehicle dynamics and the VSP modeling is also included in the processing pipeline. After exiting the pipeline, the transformed output is then fed to the feedback mechanism, which transmits specific information to the driver, according to the type of feedback used. All data is continuously stored in a local database, so that further off-line analysis may be performed. Repeated sampling from sensors will undoubtedly drain the battery in its full in a matter of hours, so there is the need of ensuring that the device is fed continuous power by connecting it to the vehicle’s internal electric circuit.
Drivers initiate interaction through the system’s main menu (see Fig. 2). In order to use the system, drivers must register themselves before receiving a 3 digit PIN code that uniquely identifies them. Vehicles registration and VSP model calibration is also required to be performed, but this may be done by the developers before the system is made available to the drivers. This will be the case when doing the experimental study with the public transit bus drivers. Besides the VSP model calibration, it is also possible to calibrate both the device accelerometer, as well set up the desired orientation of the phone inside the vehicle. This last step has some limitations, as currently we are working with a phone with only one accelerometer and no gyroscope, which limits the phone’s orientation recognition. Just before starting a trip, the driver introduces his PIN code and indicates the vehicle that he is currently using. After this the trip is marked as initiated.
In order to test the effectiveness of the feedback system, we propose using two different types of feedback: realtime and historical. In the first, we will show a real-time VSP graph that represents an approximation to the actual VSP value. The graph is an abstract representation, where it goes from green (low VSP values) to red (high values) with an approximate quadratic function increase. Additionally, actual fuel consumption, speed, and acceleration values are to be represented. In regard to the historical feedback, our system will make available two modes to the driver. The first will show the distribution of time in the pre-defined VSP bins, and the second will show a heat map of the route, indicating VSP “hot zones”. The use of historical feedback gives the driver a more broad perspective of his driving behavior, as it recalls and identifies potential patterns that may be improved. Furthermore, historical feedback will only take place when the driver is not actively driving.
In this paper we have argued that egocentric feedback on fuel efficiency can be more effective than systemic feedback on motivating driving behaviour change. Motivated by anecdotal evidence, we hypothesise that an egocentric approach is more appropriate and relevant. By re-defining the VSP surrogate metric, we are able to switch between systemic and egocentric feedback while maintaining minimal changes between our experimental conditions. Orthogonal to the manipulation of the efficiency model, we describe our interest in testing the effect of instantaneous versus historic data in the feedback system.
This work is supported by the Portuguese Foundation for Science and Technology (FCT) grant CMU-PT/ HUMACH/ 004/ 2008 (SINAIS). This work is also supported by the European Union project INTERVIR+ and the local public transit company, Horarios do Funchal, S.A.