Innovating with AI in the Automotive Sector: Are Incumbents at Risk of Stalling?
Under the spotlight with the necessary green transition to operate, theautomotive seems in full moult. We expect a lot from the connected car, autonomous driving, electric vehicle or car sharing for our future mobility.
To innovate and remain competitive, the sector could in particular rely on the big data and the technologies ofartificial intelligence (AI). Some new entrants such as Tesla seem to have integrated it well and have already successfully adopted these technologies. But what about historic companies in the sector that have to deal with the complexity of their systems in place?
Admittedly, manufacturers and suppliers have invested heavily in AI in recent years, as the project shows. Valeo.ai or Renault-Google partnerships et Stellantis-SoundHound. With what objective nevertheless? Are these approaches of radical innovation, of questioning the architecture of vehicles as we know it or rather of automating tasks and improving what already exists?
Understanding the innovation processes around data in the automotive world remains essential to ensure the sustainability of this key sector for the French economy. This is what our team of researchers, from TBS Education and Scientific Management Center (CGS) de Paris School of Mines – PSL, in recent study. It is based on an analysis of more than 46 patents from the 000 biggest players in the field as well as a campaign of interviews with the applicants of 19 patents related to AI technologies.
By trying to lift the veil on innovation practices in this sector, we show that it is indeed the "improve what already exists" option that seems to have been chosen. If it seems to make it possible to control costs in the short term and to learn step by step, this approach can however limit the innovation potential of these companies. Especially since the articulation of actors along value chains is involved and also brings its share of brakes when everyone thinks that it is up to someone else to innovate.
A careful apprenticeship of engineers
One would have thought that the integration of AI was going to be a prerequisite for the design of new vehicles. However, engineers seem to use it above all to solve problems that emerge during the last phases of product development: improving passenger comfort during vehicle testing, solving sensor problems or even, more surprisingly, negotiating with equipment manufacturers.
Companies in the sector are thus adopting a progressive and cautious approach in the integration of these technologies. They are first applied to existing driver assistance systems (ADAS) and then developed in stages. If this way of doing things is surprising, it has the merit of allowing the teams to gradually learn and adapt to the AI, while avoiding questioning the architecture of the car. This could result in a significant increase in manufacturing costs. As an expert engineer in adaptive cruise control systems :
“About 3 to 4 years ago we thought that in the coming years we would have autonomous vehicles… Today that is still not the case. At the moment we are working on the development of new functions for which we can say that there is not much disruption."
Another expert, in ADAS for braking systems, continues:
"It's not necessarily a lack of honor for AI, but… AI tends to solve problems that already exist, not problems that don't exist. The perfect autonomous vehicle is nothing more than a driver.”
Insufficient quality data
However, all this has obvious limits in terms of innovation potential. Our work reveals in particular problems of data richness. Although vehicles collect a gigantic number of them, they would need to be labeled, for example, to be usable. An autonomous vehicle expert explained to us:
"My teams have hours and hours of continuous testing, but if you want to create an algorithm for multi-directional movement, you need someone looking at the camera when that movement is happening, i.e. at each turn, to note that in the database, and it's very time consuming. »
Another obstacle lies in the technical ability to cross-reference data from different sources (visual, radar, sound, etc.) to make, for example, a decision in an algorithmic logic. These technologies are still under development.
These elements seem problematic when it comes to remaining competitive both on the global market where new players like Tesla operate, but also on the new mobility market in the face of developments, for example in flying taxis. announced for 2030. It is necessary to innovate by offering radically new functionalities or by addressing new consumer needs.
Issues that are also organizational
It is thus a question of developing the expertise of the engineers in place around the sciences of the data, and some moreover have a great desire to learn more. It is therefore not enough to develop a new entity with data scientists, but rather to ensure a gradual increase in the skills of the engineers in place. As one of the interviewees, an expert in driving autonomous vehicles, points out:
"We do not develop a patent by saying to ourselves "we are making an AI patent"".
While it is necessary to develop these skills related to data sciences, they must also be better recognized to encourage engineers to complete their prior expertise. This requires work to identify these "AI communities", beyond those specifically recruited as data scientists, and who do not necessarily identify themselves as contributors. This is also explained by a definition of AI that is sometimes restrictive, for example only restricted to the use of neural networks, when there is a wide typology of possible technologies.
Another organizational hurdle is the relationship between vehicle manufacturers and original equipment manufacturers (OEMs). Along the value chain, from subcontractors to assemblers, the actors, for the time being, seem above all to pass on the responsibility for innovation through AI. Each seems to adopt similar innovation strategies. An interviewee employed by a manufacturer explains as follows:
"It is rather the suppliers who are responsible for the development of the intelligent part of the sensor. They are the consumers of the AI methods."
The cells of experts in innovation management methods in the R&D departments of companies (Design Thinking, CK methodology, etc.) have a key role to play in breathing new life into innovation with data.
Incumbent companies in the automotive sector must therefore find a balance between exploring new possibilities and exploiting their existing skills. The current incremental approach has the merit of providing rapid results and gradually getting teams used to these new technologies. The way it is currently implemented, however, hinders the adoption of more radical approaches and the emergence of truly original technological innovations that will allow companies to remain competitive in the global market.
Quentin Plantec, Professor of Strategy & Innovation Management, TBS Education; Benoit Weill, Professor, Mining Paris; Marie-Alix Deval, teacher-researcher, Mining Paris et Sophie Hooge, Professor in Management Sciences, Mining Paris
This article is republished from The Conversation under Creative Commons license. Read theoriginal article.
