RoadAI: Reducing emissions in road construction

This challenge aims to reduce CO2 emissions from Norwegian construction machines by utilizing data-driven approaches. Skanska Norge AS provides a wealth of data from their construction sites, including GPS data, machine data, vibration data, and drone maps. The challenge focuses on using this data to optimize road construction processes for increased sustainability. Participants are tasked with developing feasible solutions to reduce idle time, optimize dump truck flow, minimize unnecessary driving, and improve driving styles. The evaluation criteria consider the novelty of the algorithms, feasibility of implementation, and the sustainable impact. The competition encourages innovative approaches to leverage data for sustainable construction practices.


Introduction
1.5% of Norwegian CO2 emission comes from construction machines [1].Is it possible to reduce this emission using insights from data?The Road AI Challenge aims to address this issue by leveraging the wealth of data collected by Skanska Norge AS from their construction sites.The participants of the challenge are asked to explore whether the data can be used to increase sustainability during construction.
The participants will be provided with data from a road construction site in Viken including GPS data from dump trucks, machine data including daily fuel consumption, and drone maps of the construction site.In addition, we have a small dataset of vibration data from a subset of dumpers, and of course public data such as weather, maps etc.
The challenge is to demonstrate how the data can be exploited for road construction to become more sustainable.Sustainability can be interpreted as directly reduced emission, but also as minimizing construction time and impact on the surroundings.The current status is that many processes are manually controlled, and could benefit from automated decision support.This can for example be obtained through: • Reduction of idle time.
• Optimal flow of dump trucks on the construction site.
• Mapping of time and fuel consumption for various stretches of roads and steepness.
• Minimization of unnecessary driving.
• Optimal driving style with minimal acceleration.
• Automated classification of road types for improved planning.
• Automated detection of load cycles (when excavators fill the dumpers).
• ... At the same time, suggested improvements must be feasible and based on data that can be collected automatically.

The available data
The main data set contains data from March, April and May in 2022.

GPS data
The GPS data is recorded from iPads in the dumpers and trucks.They report timestamp, machine ID, location, type and amount of material being moved, and where the material is being loaded and unloaded.The latter are manually recorded through the driver interacting with an app on the iPads, and hence there is some uncertainty related to when they actually record loading and unloading.The unloading is usually associated with reversing the dumper and tipping the load, and the actual location can often be inferred from the GPS track (automation of dump and load points are suggested tasks in the challenge).The data is divided into trips, where one trip contains the cycle of loading, driving the load, dumping, and driving to pick up a new load.The vehicle might return to the same loading place, or another loading location.There are two folders each containing one file per date: one for the GPS pings (trips) and one for the metadata (tripsInfo) about the trips.The metadata contains the following columns: TripLogId One unique ID per trip.Same ID as in trips data and vibration data.
DumperMachineNumber One unique number per machine.

MachineType Dumper or truck
LoadLongitude, LoadLatitude Coordinate where the machine was loaded.
DumpLongitude, DumpLatitude Coordinate where the machine was unloaded.
MassTypeMaterial Type of material transported.
Quantity Amount being transported in units of tons.
The trips data contains the following columns: TripLogId One unique ID per trip linked to ID in the metadata.
Timestamp Time of GPS ping.

Uncertainty
The real position should be within this radius of the recorded position.Units is meters.

Machine data
In addition, we provide daily reports (called AEMP, Association of Equipment Management Professionals) from a set of machines with location, odometer, fuel consumption and hours used.These are only available from Skanska-owned machines whereas the GPS data are available for all machines working on the project.The only way to match machine and GPS data is via timestamps and locations.
The machine data contains the following columns: Datetime Timestamp.
Make Brand of machine.
ID Anonymized unit id.
Latitude and Longitude Coordinates when sending data.
Hour Number of hours the machine has been in use given as days, hours, minutes, seconds.
FuelConsumed or FuelConsumedLast24 Total accumulated fuel consumption or consumed within last 24 hours.
FuelUnits or FuelUnitsLast24 Units of fuel consumption.
Odometer Accumulated distance driven.
OdometerUnits Units of distance driven.

Vibration data
For a shorter period in April 2023 we also recorded vibration data from the ipads.These are recorded at 15 Hz and contain three-dimensional vibration data as described in the Apple Core Motion documentation1 .These vibration data have not been analyzed in detail before.The transmission of vibration data takes up a significant amount of the available bandwidth to be feasible for all machines at all times.Hence, any practical use of these data will require either (pre-)processing on the iPads (computational limitations) or reduction in data to be submitted (lower frequency or fewer variables).The vibration data is given in hdf files2 .The name of the file gives the TripLogId of the data.If there are several files for one TripLogId an _number is appended to the filename.The files contain the following columns: TripLogId One unique ID per trip linked to ID in the metadata.
Timestamp Time of recording.
Attitude.{Roll,Pitch, Yaw} Roll, pitch and yaw relative to the device's reference frame given in radians.
RotationRate.{X,Y, Z} How fast the device has moved around all three axes since the last device motion sample.
UserAcceleration.{X, Y, Z} How fast the device is accelerated by the user (in contrast to acceleration by gravity) along all 3 axis of the reference frame.
TrackingEventId Can be ignored.
Heading Can be ignored

MagneticField Can be ignored
EventId Can be ignored.
RetryCount Can be ignored.
AppVersion Can be ignored.
DumperId Can be ignored.
Type Can be ignored.
Assuming the iPad to be lying flat on a dashboard and aligned with the truck (top pointing forward), the pitch corresponds to the inclination of the road, the roll should only have very small changes, and the yaw would be the direction of movement which should have very strong correlation with th heading.In practice, the ipad is mounted at whichever angle the driver sees fit and the physical interpretations will be combinations of pitch, roll and yaw.

Drone data
We also provide drone data from the unloading area where the mass is being dumped.They are created with the commercial ArcGIS Site Scan software and consist of ortho-mosaic image data (tiff file format), point clouds (LAZ data format), mesh data (slpk file format), digital terrain models (tiff image with altitude information) and digital surface model (tiff image with altitude information).These data can for example be used to develop automated progress reports.
One zip-file is provided per day of drone fly-over.

Open data
In addition, all public data can be used such as Open Streetmap or theme maps from The Norwegian Water Resources and Energy Directorate4 or weather5 .

Reasonable assumptions
Some reasonable assumptions can be made about the construction site: • Excavators have almost infinite mass to move.Very likely that they will contribute to the same task over many days.
• The long term plan is not digitally available.
• No one will use the system if you have to type a lot of extra information manually.
• In reality, decision support from the algorithms can be communicated to foremen and drivers via the iPads.
In the competition they can be shown in a notebook or similar.

Tasks
We present 4 sub-tasks: 1. Develop an algorithm to improve sustainabilitydemonstrate the potential for improved sustainability impact from using data.
2. Provide a deployment plan -describe system and data requirements as well as potential challenges for deployment.Are there any ethical issues?
3. Transparency -evaluate the transparency of the developed system including explanations for how the models were trained, interpretation of predictions, risk assessment and clear guidelines for usage.
To compete for the prize money all four tasks are mandatory.Submissions of only one sub-task are allowed, but the participant(s) will not be eligible for winning any of the prizes.The tasks encourage an exploratory and innovative approach -there are no wrong answers per se! Submission and evaluation criteria Each team should submit a visual presentation (type of video max 3 min, website, slides, flyer) and a Jupyter notebook or similar which demonstrates the algorithm.The notebook needs to be submitted both as executable code where the path to the raw data is set at the top and as a pdf file with all outputs.
The jury will evaluate the submissions based on the novelty and innovation of the algorithm, feasibility of