Image credits: Bosch

Data for Automotive Companies: New Supply Chain

By Alejandro Enríquez | Wed, 07/07/2021 - 11:24

Data science is accelerating technological transformation across different sectors and the automotive industry is not the exception. From business analytics to understand consumer trends to machine learning models for autonomous vehicle systems, data has become an essential part of the automotive value chain. 

Defining the Data Supply Chain

Artificial intelligence and machine learning models use software(machine)'s ability to learn for itself new processes to build a prediction and often take an action based on it. For instance, out of a series of images presented, applications can detect which of those pictures have a person on them. The very first step for building a machine learning model is data collection. Data is the raw material that will later feed the predictive model. The second step is to prepare the data. Often, data is not correctly labeled and needs to be 'fixed' before introducing it to the model. The third step is to either design or apply an existing machine learning model that will process the data to later train that model so it performs the desired action as effectively as possible through each iteration. The final step is then starting to make predictions and execute a certain set of protocols.

These processes are usually considered a human-less process. However, academics from Oxford Internet Institute (OII) – one of the world's leading research institutes on data and society – highlight that "AI requires global supply chains and a wide range of workers, many in the Global South who increasingly do routine and routinized work to ensure that AI systems function," says the Artificial Intelligence in the Workplace Report.

Fellows from the OII presented a wide variety of case studies regarding AI applications across different industries and countries. "The aim of this report is to present a more comprehensive dialogue around the use of AI as more workplaces roll out new kinds of AI-enabled systems," stated the report. The case studies presented involved more than 400 news, academic and industry reports between January 2019 and May 2020. OII found that in some cases, AI requires a vast amount of work. "At Google, the team responsible for the data sets that make the company’s voice-activated Assistant work, Pygmalion, relies on the “painstaking” labor of annotating datasets by hand. Much of this work is done by temp workers,” reads the case study. "AI is marketed as automatic but it is often based on tedious behind-the-scene work," highlighted OII. "Many companies are still heavily reliant on invisible human labor, often based offshore in India or China."

Why Is Machine Learning Important for the Automotive Industry?

Machine learning and AI applications are strongly linked to the evolution of the automotive industry. Autonomous vehicles essentially rely on artificial intelligence and machine learning for its development and depending on the level of autonomy, more complex algorithms are required for the vehicle to adequately respond to a given situation. At the moment, according to the US NHTSA, there are no available Level 4 or Level 5 autonomous models in the market.

This level of dependence on machine learning also explains the importance of semiconductors for the automotive industry. IHS Markit's notes the strong relationship between these technologies: "Autonomous driving is one of the key application areas of artificial intelligence (AI). Autonomous vehicles (AV) are equipped with multiple sensors, such as cameras, radars and lidar, which help them to better understand their surroundings. These sensors generate a massive amount of data. To make sense of this information, AVs need supercomputer-like, nearly instant processing capabilities. Companies developing AV systems rely heavily on AI, in the form of machine learning and deep learning, to process the vast amount of data efficiently and to train and validate their autonomous driving systems."

Autonomous vehicles are also strongly linked to shared mobility. Large ride-hailing companies including Uber and Didi have launched pilot tests with autonomous vehicles in certain areas. In the case of DiDi, the company recently announced a comprehensive partnership with Volvo Cars for the development of Level 4 autonomous vehicles for DiDi's ride-hailing platform in China. Volvo will provide XC90 models that will be equipped with all the necessary backup systems for autonomous functionalities, while collaborating with DiDi to integrate additional software and hardware.

Data: Part of the Automotive Value Chain

To analyze and produce an successful outcome to a problem presented to a vehicle, models need to collect data and prepare it for analysis. This is the often 'unclear' and labor-intensive side of AI supply chains that OII noted. "The outsourcing of manufacturing was a more visible process, rendered through things like ‘Made in China’ tags in clothing. The AI supply chain ought to have a similar ‘Made in’ attribution scheme to better understand the global assembly of a technology often considered to be purely technical in nature," notes the OII report.

Who are the main Tier 1 data suppliers? There are two major companies that are closely involved with the development of ADS. publicly displays its collaborations with ZF, Siemens, Daimler, DELL and Continental AG. Another big name in the segment is, that works with Volkswagen and Bosch. "Highly accurate annotation is an indispensable prerequistie for supervised machine learning. We rely on the labeling service and tools from" said Florian Faion, Research Scientist LiDAR Perception at Bosch, on the company's website.

Faion refers to the work of labeling data. We mentioned before that companies need raw data, which in the case of ADS usually refers to the vast amount of information collected by vehicle cameras and sensors. The machine cannot recognize objects by itself. It needs to be trained to recognize a person, a bicycle, another vehicle, road lanes and traffic lights, among other elements. What companies like and do is to provide teams that facilitate the labeling of each object in videos or pictures. These teams manually label a vehicle as a vehicle and a person as a person. This is what OII referred to as "tedious tasks" performed by individuals behind AI models. "Our ML-assisted tools and user-friendly interface ensure that your annotators are more productive, creating labeled datasets five times faster," states’s website.

The Tier 2 layer of the data supply chain involves companies that supply the large volumes of data needed, including videos or pictures from vehicle cameras or LiDARs. This layer is more pulverized than the previous one as data companies and sensor manufacturers are involved in the process.

Photo by:   Bosch
Alejandro Enríquez Alejandro Enríquez Journalist and Industry Analyst