YND TeamNovember 20, 2018
Initial Research
As frequent users of car sharing in Berlin, Munich and London, the team was talking about a common problem we had all experienced: messy and dirty shared car interiors.
We started looking into how much of an issue this was for other users, and we found a surprising number of unhappy comments posted on social media.
The team decided to do more research offline to see how much of a problem dirty and messy shared cars really are. So we hit the pavement in our home city of Berlin to see for ourselves. The team explored five parts of the city to check cars on different days and times. The aggregator app Urbi made it easy to find available car2go, DriveNow and driveby cars nearby.
We found that lots of cars were clean, but plenty were dirty or contained other interior issues. Overall we found that 38% of 85 car interiors in Berlin were unpleasant, messy or very dirty.
For more on the research, you can read more in our Medium article.
The Problem for Operators
We also looked into the service delivery side, and realised that car sharing operators don’t have real time data about the cleanliness of their fleets. Since checking the condition of vehicles is currently a manual process, cleaning them is inefficient.
Operators clean vehicles only semi-regularly (on average every 10-15 days or more) and a lot can go wrong in that time. By the time issues are identified it’s too late for the customers who have had a bad experience of the service and poor impression of the vehicle brand.
The solution
After discussing the problem we thought we could solve it using Computer Vision and AI.
The idea is deceptively simple: cameras installed in shared vehicles take images once the ride is finished and the car is locked - this protects the user’s privacy by ensuring they are not recorded. Algorithms trained using thousands of images then ‘scan’ the images to identify dirt, rubbish, belongings and other issues like broken glass that shouldn’t be there.
The operator then receives an alert when issues are identified by the system and can respond accordingly e.g. by taking the vehicle offline temporarily and dispatching a cleaning team.
Here’s a simple overview of the system:
Over time the system becomes more sophisticated and can make decisions with minimal or no human intervention.
We are also interested in adding other sensors to the modular platform that CleanAI is based on - IRIS. These could include sensors for identifying chemicals and harmful particles in the air that could pose a safety hazard to commercial vehicle users, such as delivery drivers and construction workers.
VW Challenge
Validating the CleanAI concept, Volkswagen selected the team as finalists among 70 applicants from 16 countries in the VW Nutzfahrzeuge Innovation Challenge 2018.
Alistair Cadman, Chief Product Officer, and Michael Dominic K., Chief Technology Officer, pitching to the VW Challenge panel at the DRIVE VW Forum Berlin.
While the team ultimately didn’t take away the win, we made excellent contacts in the German mobility services startup community, including the winners and our new friends at Berlin startup Chargery - congratulations to the Chargery team!
Up Next
The CleanAI team are currently discussing pilot opportunities with potential partners. The next step is to install the existing technical proof of concept in one or more shared vehicles for real world testing and training of the algorithms on an expanded item set.
You can follow CleanAI’s progress via Twitter and Medium and in the meantime you can also reach out to YND about any AI-based projects or questions you might have!
This post was written by Alistair Cadman. In need of some brain power? Reach out to us via hello@ynd.co with your questions about ML/AI and mobility projects.