Skip to main content
CORDIS - Forschungsergebnisse der EU
CORDIS
CORDIS Web 30th anniversary CORDIS Web 30th anniversary

SOUTHPARK - SOcial and Universal Technology HelPing to detect ARrivals via sdK

Periodic Reporting for period 3 - SOUTHPARK (SOUTHPARK - SOcial and Universal Technology HelPing to detect ARrivals via sdK)

Berichtszeitraum: 2017-08-01 bis 2018-07-31

predict.io has been set up as a company to address individual, as well as societal problems related to smart mobility by building a smart, integrated, technology-based, easy to use and cost-free system that can easily be integrated in all kinds of mobility solutions. The outcome of the project will be an ensemble of algorithms that had been tested on large scale in the operational environment of different European cities.

The SOUTHPARK project is set up to fulfill two overall objectives:

First, it will enable innovative mobility solutions in Europe’s metropolitan areas. Predict.io will integrate the technology with its automated start and stop detection in different mobility apps. This will be achieved when the SDK generates 1 million data points a day for real-time applications as well as business analytics. We will follow the already developed commercialisation plan with the predict.io sales approach to explore and pursue business opportunities in the mobility sector and recruit a large-scale sample of data points. The goal is to enable more convenient, safer, environmentally friendlier, and efficient mobility solutions.

The objective has been met. The SDK and analytics suite has been successfully launched across geographies and is currently producing ~75 Million data points per day. The target to produce 1 Million data points per day has been far exceeded. Test users like the City of Barcelona are benefiting from the technology and are evaluating long term use.

Second, the SOUTHPARK project will enhance the existing predict.io algorithms with a component that enables the technology to self-adopt to new local settings. In order to achieve this objective predict.io will repeatedly collect and analyse locally deviating data sources in order to identify, attune and optimise the locally diverging features for its STOP detection system. Thereby it will train and enhance the predict.io algorithms by testing and cross-testing large-scale datasets in the end generating a self-adopting component that can adjust itself to any type of local mobility situation. Fulfilling this objective will reduce the adaptation costs and time to localise to new local settings by estimated 75%.

We have successfully built a system that can automatically learn from user input which significantly lowered the cost for expansion to new territories. Furthermore this laid the foundations for our rapid testing of new verticals. Today we are able to produce ground truth data from live users in order to feed the inference algorithms with patterns to search for. Regardless which type of inference we aim at (parking spot, transport mode, venue, visit intent, etc.) the performance is measurable and changes to the algorithms can be done on the fly. Before launch of the system building a new inference took us between 3 and 6 months. Now we are able to train the system within 1 week with user input. Hence, objective 2 has also been met very successfully.
Work package 1 - Project Management
All administrative tasks have been progressing well and all deliverables due during this reporting period were timely delivered. The project internal communication is excellent, with monthly team meetings to monitor progress of tasks and to manage any issues that occur. WP1 organised the coaching sessions provided by the EASME.

Work package 2 - Localisation and scale-up studies
To localise the predict.io system throughout cities in Europe, WP2 worked on the recruitment of test users and tested the functioning of the developed technology. The predict.io sales team undertook great efforts to reach out to public and private partnerships to provide a data base that would be satisfying for consumers. Several different prototypes were developed and tested constantly.

Work package 3 - Machine learning
All data gathered by the predict.io technology provide the basis for the building, training and testing of the algorithms. In WP 3, data scientist and machine learning experts work on the collection and evaluation of the various data sets, the extraction of possible features for the technology and the training and implementation of the algorithms. A demonstrator app of the Start/Stop detection has been developed as well as and application called "waypoints" that is able to not only detect and predict the start and end point of a trip, but also waypoints in between.

Work package 4 - Dissemination
SOUTHPARK focused in the second project year again on participating at events and applying (successfully) for awards. Besides, communication on social media has been ongoing with a focus on B2B clients. A blog section has been established on predict.io's website and organic press coverage could be generated.
Concerning objective O1, the predict.io system produces around 250,000 events per day by the end of the first reporting period. Localisation proves to be complex as all system’s different functions require separate components to be adapted. Still, the adapted product design and use cases show reduced dramatically localisation cost in conjunction with better marketability.

Focusing on the detection and prediction of the start and the stop of a user’s journey, objective O2 is progressing well despite of many different approaches that had to be followed in the development of the product in order to make the adoption of the algorithms easier, faster and more reliable.
Predict.io's CEO Silvan Rath presenting at CUBE tech fair 2017
Example of the latest predict.io demo app
Example of the latest predict.io demo app
Example of the latest predict.io demo app