Machine learning for everyone

Machine learning for everyone

Picture the scene: an open office space has been transformed into a racing ring, you hear music and the steady chatter of excited colleagues, and all you can see are smiles and groups of people operating miniature cars with concentration and focus etched on their faces. It was not a typical Tuesday at the KONE premises in Hyvinkää, Finland.

Published Apr-16-2024

It was the day of the epic Amazon Web Services’ (AWS) AWS DeepRacer event, where participants get to experience and make discoveries about machine learning (ML). And what are we talking about, exactly? In short, a competition with a mission to use machine learning to teach a miniature car to navigate through a physical track as fast as possible. 

Over 100 of our colleagues from different teams were eager to jump in, building and training race car models. Virtually at first, and then testing them on a real track at our Hyvinkää site – no previous tech or coding experience needed.

What initially started as a grassroots initiative for machine learning, turned into a fun event with participants from various units joining in, fostering collaboration and digital competence building in an engaging environment. 

Some of the teams had even been preparing for the race over the weekend – not because they had to, or due to a deadline – but because it was genuinely fun and interesting. This is just one example of how we drive industry transformation with the help of digital innovations.


In the virtual world, everything is flawless – but how does reality stack up to it?

It all started with Petri Rosenström, KONE’s lead architect for the Cloud Platform. “I’ve organized many events where people have been able to learn and improve their technology skills. When I heard about the first DeepRacer event in Finland, I wanted to get it to KONE as soon as possible,” Petri says. “The idea in the background – to experience machine learning in a fun and collaborative way – was brought to life thanks to our strategic partner AWS, offering multiple tutorials, courses and events around cloud technologies and artificial intelligence (AI),” he concludes.


The type of machine learning specifically explored at the DeepRacer event is called ‘reinforcement learning’. This model is similar to the methodology used in teaching dogs or any other animals, even humans. It is about learning new behaviors for rewards – in other words, learning to take actions with good outcomes and avoiding actions with poor outcomes. One of the findings from the event was that it is easier for humans than machines to apply the theory in practice.

Let’s hear from our winners – what was the key to success?

Meet our racing champions who reached pole position; Rui Heinonen and Sudarshan Koirala, both from the Enterprise Data and Analytics Platforms team at KONE IT.


Rui Heinonen started as a trainee in the Analytics team. He then continued to do his master's thesis at KONE, concentrating on dynamic maintenance planning. For Rui, working in the technology field wasn't something he always envisioned, but during his third year of studies at the Aalto University School of Business, he discovered an interest in coding and mathematics. As a data scientist, Rui has worked with pricing analytics and currently collaborates closely with developers and business stakeholders on sales and marketing analytics.


Sudarshan Koirala also started at KONE as a trainee data scientist, then moved on to the role of associate data scientist and later to machine learning engineer. After taking an interest to physics and mathematics in high school, Sudarshan switched to business and then to software engineering, before finally jumping into data science. Currently, in addition to being a machine learning engineer and helping business stakeholders with a churn prediction solution called Risk Ranking, Sudarshan is also the Scrum Master for one of the teams in the Enterprise Data & Analytics IT Department. The team focuses on several AI, ML and advanced analytics applications related to marketing, sales and tendering, together with business stakeholders.

“Typically, data and analytics projects at KONE are done in close cooperation between IT and business teams and entail a wide range of tasks – from identifying business objectives to gathering relevant data, cleaning and pre-processing the data, performing exploratory analysis, utilizing training and serving models with statistical or machine learning techniques, and finally communicating the findings to stakeholders for informed decision-making,” Sudarshan explains. “Depending on the stage of the project, we work with different technologies, such as Databricks, AWS, Visual Studio Code,” Rui says.

“A key takeaway from the DeepRacer competition, in addition to the basics of reinforcement learning, was to not be too confident and learn from mistakes. That’s exactly what we did and I think it eventually helped us win the race – we didn’t only train our model cars, our strategy was to consider and experiment with different things and use a common sense approach, while also observing what others were doing,“ Sudarshan Koirala explains. “Even though we work daily with machine learning and data science in our projects, I would not say we had an advantage compared to colleagues from other domains. Reinforcement learning was new for both of us,“ he continues.

From left to right: second place winners - Tero Hottinen, Terhi Istomin, Niina Kontio, CTO of KONE Tomio Pihkala, first place winners – Rui Heinonen, Sudarshan Koirala

Rui Heinonen highlights the importance of what he dubbed the “common sense approach”: “Rather than just immediately plunging into coding, our strategy was to first research the dynamics of the track and the race car. This involved studying the track layout, angles, and determining optimal speed and acceleration for each section of the track to achieve a certain theoretical lap time.”

“We enjoyed the AWS DeepRacer competition, it was a good way to get started with reinforcement learning. It also provided an excellent opportunity to observe how models behave differently in the real world compared to the virtual environment,” comments Rui. “Moreover, the event gave us a chance to network and build connections with colleagues beyond our typical work circles, all within a relaxed setting,” Sudarshan adds.

Learning while having fun unleashes true teamwork and creativity

Many participants shared the sentiment that the DeepRacer event was filled with great energy, fun and excitement.

For example, Risto Jokinen, senior lead SW Developer, Drives, SW Architecture and SW Engineering, commented as follows: “I came to KONE at the age of 15 and I’m still learning”.

Charandeep Matta, chapter lead, IoT cloud platform, said: “Oh man, you had to be there to realize what’s going on! The competition showed the potential of AI and the randomness that exists, it opened my eyes to a different world of possibilities”.

Watch the short video below to get a feeling of what the day was like.

“I am proud to say that KONE is actually only the second company in Finland to experiment with this fun learning event. I’m also happy that not only KTI experts joined – we had racers from finance, HR and communications, too! This really proves that everyone at KONE can be a digital ambassador on our journey to improve the flow of urban life,” sums up Tomio Pihkala, EVP Technology & Innovation at KONE.

”We are happy to co-innovate with the KONE experts and contribute to experimenting with new ways of learning. Organizing the DeepRacer event together was a great example of how KONE wants to stay at the forefront of digital innovation,” says Sari Uusitalo, Country Manager AWS Finland.

We’re excited to see where our partnership with Amazon Web Services (AWS) will bring us on our quest for continuous learning and co-creation. Come along and join the ride!