Team #15 // Good to Go

Executive Summary

As the popularity of e-scooters in larger cities is increasing, so is the rate of accidents they cause. The e-scooters were allowed into operation in Copenhagen in January this year, and by today, the number of serious accidents has surpassed 100. A third of them is estimated to be due to drunk driving, and the operators are searching for ways to solve this problem. Responsible Mobility integrates a AI-software into the mobile applications of the operators that simply blocks the access to the service for drunk users. The first step of realising the concept is to test the solution on a smaller scale in Vejle Kommune. In the long run, the solution is scalable to the general shared mobility market of the big metropolitan areas.

Problem

E-scooters are one of the most increasing trends in sharing transportation in the big urban areas of the Western part of the world. In Copenhagen, 6 operators and approximately 3500 scooters have already moved into operation since the legislation allowing the e-scooters was passed in January, 2019. Unfortunately, accident rates associated with the e-scooters are high. Copenhagen just recently clocked the 100th serious accident involving e-scooters. Riding an e-scooter while being drunk is as illegal as driving in a car while being drunk but nevertheless, it is estimated that a third of the serious accidents are due to people driving under the influence of alcohol.

Present course of action and existing solutions

To combat this issue, some operators shut down the service of the e-scooters by night, either by collecting them to charge or by remotely shutting down the system. Collecting the e-scooters represent unwanted costs and the long periods of inactivity also results in sunk costs for the operators. Presently, the only applied method for detecting drunkenness is using hardware such as breathalysers. So-called “alkolåse” that requires a breathalyser-test are today fitted in some cars and trucks. But to implement a hardware-based solution on an outdoor public transport vehicle such as the e-scooters would be costly and not very durable.

Good To Go: The solution

Good to go delivers an AI-based software plug-in for the e-scooter operators’ mobile applications. With our software in operation, in certain high risk periods of time such as Friday and Saturday nights, the e-scooter users will have to pass a drunk test to access the booking system of the operator. The test requires the users to record 5 seconds of audio and video, where they speak a statement out loud. The software evaluates the voice and the video of the face to determine if the user is drunk. In a paper proposed by Metha et al. 2019 [1], known and open-source deep learning models trained on Youtube-videos can detect drunkenness through a combination of audio and video with an accuracy of approximately 88%. Thus, we aim to train the model on open-source data and then tune it with test data gathered from real-life situations to make the model fit our case.

Challenges to handle with the solution

When “Good to go” is implemented, a few dillemas can arise . If the software detects wrong, that is f.ex. that the person is actually sober but estimated drunk, the person would like to have another try to hopefully pass the test the second time. While this is a realistic scenario and we don’t aim to block sober riders, it could also be the case that the person (rightfully) fails the first time. This drunk person will also be likely to want another try on the drunk test. While the final decision will be dependent on the stability of the model, we have some suggestions on how we could handle these scenarios:

1) Block the user’s that fails the test from the app for 2 hours, making sure that no drunk people pass the test the second time

2) Allow a second try, but extend the drunk test to 10 seconds so that the model will have 15 seconds in total. Hopefully will have so much data on the person that the model is more likely to let the sober person pass the second time, while the drunk person still fails.

The measures taken toward these challenges are also dependent on how ‘strict’ the software should be. Is the goal to avoid that we block some sober riders from using the e-scooters, or is it to make sure as many drunk riders is blocked using the e-scooter? This decision will probably be made in collaboration with the operators and the municipalities.

Challenges with the model

When it comes to the AI-model it self, there will also be some hurdles. A ‘classic’ machine learning/AI problem could be if the data is bad. F.ex. that the lightning conditions are bad or the video is recorded in a noisy environment. Therefore we will probably have to do some preprocessing of the data before it is fed to the model. It is impossible to say exactly which problems will arise during the process of building the model, but with the results we have seen with the newest neural networks we are confident that this can be done.

Customers and stakeholders

Good to go’s direct customers are the e-scooter operators. They are the main owners of the problem and therefore suffers the most from it as well. Besides the real and sunk cost the operators sustain in preventing the accidents, people are likely to stop using a product if they have bad experiences with it, and having an accident is quite a seriously bad experience. Along with that, the rising amount of accidents are very hurtful to the public image of the operators, as their products are associated with low safety and bad management.

The citizens are the end users of the solution and are obviously a hugely important stakeholder. Ultimately, our solution will protect them from the high risk of suffering severe injuries while drunk riding the e-scooters. On the other hand, the solution needs to be as smoothly integrated as possible, so the sober users still feels it is easy to access the e-scooters.

The last very important stakeholder in this case is the municipality, ultimately controlling the regulation of the e-scooters in their cities. Their interest is in both providing their citizens the best and most convenient means of transportation while governing their health and safety, and therefore our solution appeals to both their main interests concerning this problem.

Five unique value propositions

What makes your solution/concept unique?

  1. Drastically reducing the number of accidents on e-scooters related to drunkenness
  2. Contributing to a better public image of e-scooter operators
  3. Helping e-scooter operators and municipalities to take a stand against drunk driving.
  4. Creating awareness on the topic of drunk driving by giving it presence in the life on the streets.
  5. Making people feel safer in the streets.


Financials (Business model)

We offer e-scooter suppliers an integratable software for their applications to eliminate both real and sunk cost associated to the accidents and the action currently taken to prevent them. Furthermore, a better public image is valuable and crucial for the continuous growth of the e-scooter concept. This justifies our revenues, which are generated by a monthly license payment for the software including support and continuous development.

Impact and scalability

The main impact of our solution is to make green mobility in the cities safer, contributing to SDG 3.6 and 11.2. The scope of the concept has a wider future potential. Every shared vehicle that opens on a mobile application can integrate our solution into their apps. Especially car sharing cars is an exciting prospect, since car accidents through drunkenness are by far the most severe.

Thus, we envision three strategical stages for our solution:

  1. Short term – the next step is to test and develop the software in collaboration with Vejle Kommune and their e-scooter supplier, Voi. Jette Vindum is in charge of the deployment of e-scooters in the city of Vejle, and has agreed to make Vejle Kommune a testing platform for our solution.
  2. Medium term – aiming for all e-scooter operators
  3. Long term – aiming for all means of shared transportation


Team

Johannes Boe Reiche (BSc Artificial Intelligence and Data Science, ongoing)
“The prospect of working with some talented people with different skill sets was a great motivation for participating in Oi-X. Furthermore, I was interested to see if my skills in data science and AI could be a part of the solution, and now it seems we have identified a problem and a solution that fits my interests perfectly.”

Sofus Dynes Steenberger (BSc Strategic Analysis and Systems Design, ongoing)
“My main motivation with my studies is to be able to make a difference in turning the world sustainable. The Oi-X provides a good platform to work with specific and challenging sustainable cases, boosting my development and ideation skillset, which I hope to really use to fulfill my ambitions. My education has provided me with a broad skill set in both data science, economics, strategy and decision making and I think the problem we have decided to solve is complex and exciting and requires more than just a technical fix.”

Villads Stokbro (BSc Artificial Intelligence and Data Science, ongoing)
“I would like to apply my AI-skills to real world problems and work intensely through a weekend to get an idea for a potential startup.”

Valerie Grappendorf (BA Internet of Things – Design of networked systems)

“My motivation for joining the nordic health hackathon was that I am currently working for the company Corti whose vision is to solve the problem of deaths caused by misdiagnosis with AI and machine learning and I now wanted to explore other problems/solutions myself. My skills lay in design-thinking, product development, user experience and visual design.”

Group Members

List full name and e-mails

Share

Share on facebook
Share on twitter

More related