AI in autonomous cars

Árpád Takács works at AI Motive, the full-stack driverless car software developer.

Dr Árpád Takács works as an Outreach Scientist for AI Motive, one of the leading lights in development of new techniques to help driverless vehicles understand their surroundings and start thinking for themselves. The company has offices in Budapest, Helsinki and Mountain View, California.

He’ll be running the ½ day workshop (part of our 1 day event) entitled Artificial Intelligence in the ecosystem of self-driving at our upcoming event at OAMTC Teesdorf, near Vienna Austria, on 28 July – so we caught up with him to find out more.

 Why is AI so important for driverless cars?

Today, most of the ADAS (advanced driver assistance systems) are relying on classical computer algorithms, especially in the vision domain. This is very suitable for simple, independent tasks, such as lane detection, forward collision warning, even simple decision making.

As the number of ADAS functionalities increases, the simultaneous detection, interpretation of the environment gets more complex, requiring a large number of hand-written rules and methods for solving the task of self-driving.

The power of AI is that it is a scalable approach, AI-based methods rely on a training data where visual or behavioral features are learned automatically (i.e. we don’t have to tell the AI what features to look for on an images), and most importantly, AI can generalize much better than classical algorithms, increasing its robustness.

Can you quickly explain the difference between AI, Machine Learning and Deep Learning?

These phrases mean different things to experts, and the definition of AI is a difficult task even today.

In my interpretation, AI can be a machine, an application or an artificially created consiousness, reflecting cognitive intelligence.

Machine Learning is a tool, which provides th AI the ability to change a behaviour intentionally and in a reproducable way – Machine Learning is the ‘training method’ of AI.

Deep Learning is closely assiciated with Neural Networks, one of the most promising methods of AI, where Deep Learning is hierarchical, structured learning method for Neural Networks with multiple hidden layers.


Getting started with artificial intelligence is daunting – any tips for a first-time software developer on how to get started?

One of the best ways to start understanding AI is through vision-based detection tasks, where image recognition (particularly digit recognition) is a well-documented, popular area. Starting with the basics of machine learning, SVMs, simple classification tasks, then moving towards Neural Networks, Deep Learning and evolutionary algorithms is a standard way to go.

Software frameworks and public datasets can help with the training, such as using Caffe, Theano or Tensorflow.

[Video courtesy of NVidia]

What do you think about the Self Driving Track Days project – is community outreach important?

Self-driving is one of the hottest topics of our times technology-wise. This is going to be the first technology where AI will be used in a safety-critical system on a world-wide basis, and it is crucial that the public understands the driving force and brain behind these machines.

So yes, community outreach is a key to bringing self-driving cars on the streets. Self Driving Track Days is a great opportunity for this.

AIMotive have permits to test autonomous vehicles in Hungary, Finland and California

You’re recruiting – what qualifications and experience are you looking for?

In general, we are looking for qualified, agile professionals with excellent programming skills, string mathematical background (for AI researchers especially) and people with experience in working with automotive development.

More precisely, we are looking for AI and image processing researchers, C++ developers, software and embedded system engineers, control engineers.

Is finding enough good recruits challenging?

It has always been challenging since the beginning – as a small company, we need professionals with a wide field of knowledge, out-of-the-box thinking and the ability to switch focus in a short time.

While finding such recruits is not easy the problems around self-driving are so exciting and challenging that it really helps us finding the brightest minds.

What advice would you give to automotive technology companies that are worried about how they fit into the next generation of cars – the new driverless ecosystem?

The automotive industry started going under a major change in the last years, software is taking over the most important place, and the position of OEMs as integrators is less stable than it used to be.

Most of the OEMs and Tier 1s have recognized this and started with the development in-house, but now there is a lot of space for new companies and technologies in the new ecosystem (such as self-driving software, connectivity etc.).

The industry is opening up, so my advice for worrying companies is to take this as an opportunity to start developing new technologies together with the newcomers – OEMs and Tier 1s have the experience in productizing an idea, but this idea can now come from someone outside of the traditional automotive field.

With more than 50 attendees registered from more than 30 companies, time is running out to book your place!

Join us for a full day of training and workshops, no prior learning required – ideal for people wanting to learn more about driverless vehicle technology!