Meet STELLANTIS
CRF (Centro Ricerche Fiat), located in Orbassano and Trento (North of Italy), is part of STELLANTIS Group (STLA), focused on research activities: its mission is to develop and transfer innovative powertrains, vehicle systems and features, materials, processes and methodologies together with innovation expertise in order to improve the competitiveness of STLA products, to represent STLA in European collaborative research programs, joining pre-competitive projects and promoting networking actions, and to support STLA in the protection and enhancement of intellectual property. Within this context, CRF is actively involved in key European Projects, such as Hi-Drive, EVENTS and Podium.
What are some of the biggest challenges the automotive industry faces in adopting advanced software and technologies in vehicles?
Since this question may become extremely broad if we consider all the SW and technologies in vehicles, I will focus here on some aspects, particularly on the applications of Artificial Intelligence (AI) on-board vehicles.
As well known, with the development of AI, automatic driving (AD) technology has been gradually deployed to relieve drivers from tedious driving tasks and to improve vehicle driving safety. In particular, AI is progressing rapidly, and companies are focusing on the development of generalist AI systems that can autonomously act and pursue some goals. Of course, this can increase the capabilities and autonomy of AI-based systems, thus amplifying their impact, but at the same time it can lead to some risks, including large-scale social harms, malicious uses, and – above all – a loss of human control over autonomous AI systems, as well as a loss of expertise and capability (crucial aspects, in driving).
In fact, AI systems have historically been designed, developed, and deployed with a technology-centered focus. In order to reduce the risks mentioned before – mainly related to misuse and negative societal impacts – new frameworks have been proposed. One of them, which I regard as particularly relevant and promising – is the human-centered AI (HCAI), in which the strength of both the human and the artificial agents can be used for a safe(r) driving task.
As cars become increasingly connected and reliant on software, what obstacles must companies overcome to ensure a reliable and user-friendly experience for customers?
Following what mentioned in the previous point, we highlight here the concept of human-machine co-driving, which is a driving process, where both human and machine agents complete the driving tasks together. The goal is to achieve a smooth transition from automatic driving to manual driving (and vice-versa, if necessary). In this context, both human driver and automatic system remain in the control loop according to their respective task distribution and allocation, as well as control intentions. This means that the human-machine shared control achieves the dynamic allocation of driving authorities. Clearly, the intention consistency and the driving authority allocation strategy of human-machine system will affect vehicle driving safety. Therefore, how to allocate control authority (taking into account the human-machine intentions and the safety aspects) is a key challenge (even considering that can be an intractable problem), not yet solved and thus worth of further research and investigation.
In driving automation of level 1 up to level 3, as defined by the Society of Automotive Engineers (SAE) International, there is a human-machine co-driving process that “implements” the transition between automatic driving and manual driving: the human-machine shared control strategy offers a promising solution for this issue, allowing both automatic systems and human drivers to have control over the vehicle, with authority dynamically allocated. This strategy can compensate the limitations of automatic system while also reducing driver fatigue.
What are the key challenges for automated vehicles in accurately perceiving their surroundings and making safe decisions in real-time?
Autonomous Vehicles (AV) represent an interesting solution to offer safer and more efficient mobility. However, challenges in traffic safety and driving policies persist, particularly in complex scenarios (such as, executing lane changes in high traffic, turns in mixed and varying urban environment, and so on). In this context, the safety of AVs primary depends on their ability to handle complex, near-collision scenarios. Here there are a number of challenges. First, AVs depend on sensors (e.g., Cameras, RADAR, LiDAR, etc.) for environmental perception, but poor perception can be very dangerous, resulting in severe accidents, or anyway highly risky situations (with possible damages). Second, the performances of AVs need to be accurately assessed. However, due to the scarcity of high-quality data in the real world, comprehensive and quantitative assessment of AD is not always guaranteed, so it remains challenging. Traditional AV testing requires driving hundreds of millions of miles in natural environments to cover these high-dimensional and rare critical scenarios, but this is extremely time-consuming, costly and sometimes even unfeasible (by the way, these problems increase the difficulty of model training).
Effective solutions to these two main issues can be represented, respectively, by accurate and advanced algorithms for perception and decision making, as well as the extensive use of SW simulation, to create challenging safety-critical scenarios in an artificial environment rather (maybe in combination with real-world data).
Please provide a closing remark related to EVENTS project
With reference to the points 3) and 4), EVENTS project fully addresses these aspects, with its two main objectives: i) design and integrate a dedicated perception platform to enhance the environment reconstruction in complex scenarios (particularly, adverse weather conditions and urban context); ii) develop and implement advanced decision-making algorithms, to improve the decision capacity of the automatic system, or even to offer a supporting system to the driver that has to take the right decision depending on the specific situation.
The modules developed by the project are tested first in the driving simulator and then integrated and evaluated on real prototype-vehicles (the project demonstrators).
Finally, last but not least, since AI is a set of techniques data-based, it is fundamental to take into account how the data are collected and selected, considering also aspects such as the privacy of people and the cyber-security of these personal data.
The above feedback is provided by Mr. Fabio Tango, Senior Researcher within STELLANTIS. The views and opinions expressed are those of the author(s) and do not necessarily reflect those of STELLANTIS.
