I’m proud to be able to publish an exclusive interview with Alberto Broggi, General Manager at VisLab. This Italian company specializes in autonomous driving technologies and is now part of Ambarella. Alberto shares valuable insights into the development of autonomous driving technologies, the close relationship between hardware and software in autonomous vehicles, and the industry’s future direction. He also discusses the potential models of autonomous car utilization, the evolution of sensor technologies, and the significant impact these technologies could have on other sectors. Whether you’re an industry professional, investor, or technology enthusiast, this interview offers a rare and compelling look at the exciting future of autonomous vehicle technology.
1. Alberto, you are well-known in the autonomous vehicles industry and research fields. For the few who don’t know you, please tell us your position, role, and how you got there.
I’m currently General Manager at VisLab, an Italian company working on autonomous driving systems. During my research years at the University of Parma, I started a group on autonomous driving when very few people were working in this field in the world. We’re talking about the 90’s. We developed the first prototype of an autonomous vehicle and tested it on the 6-day tour (‘MilleMiglia in Automatico‘) throughout Italy in 1998, where the vehicle drove itself in semi-autonomous mode for the whole 2000 km. Semi-autonomous means the driver sets the speed while the system handles the steering. A few years later, we participated in the DARPA Challenges, and in 2009, together with my students, we decided to create a spin-off of the University; our research in autonomous driving culminated with a 13.000 km autonomous driving test from Parma to Shanghai, China, in 2010, which marks an important milestone since we recorded all the trip and started treating it like ‘big data’ for subsequent improvements of our system.
In 2015, VisLab was acquired by Silicon Valley company Ambarella, a chip maker specialized in ultra HD video processing and compression, with the aim of producing chips for autonomous driving. Together with Ambarella, we produced a number of chips of increasing capabilities of handling some functions of AD, while last year we presented our main chip to be used as domain controlled in autonomous cars. This chip handles all the blocks of functions, from perception to data fusion and map generation to planning, used for AD. A very important aspect of this chip is that it accelerates all the AI functions required not only by the automotive domain but can be used as an AI accelerator in many other fields.
2. Will the separation between hardware manufacturers and software developers prevail over the Tesla model based on in-house development and deep integration between hardware and software?
My own view is that it certainly will, as it happened when other technologies got mature and widespread; however, it will take some time to get there. We are now still in the development phase, and I like to remember that currently, there is not a single solution -not even for the sensor suite- for autonomous vehicle technology. As it happened in other fields, and in the example you mentioned in your question, the first solutions are developed with a deep vertical knowledge of the whole system, starting from the silicon up to sensors and processing, till the final AD functions. Deep integration provides high efficiency and capability to make the best use of the hardware, sensors, and vehicle integration. However, I would not be surprised if, in the next future, when sw and hw will be thoroughly tested and clearly mature, it will be possible to trade some inefficiencies for a larger benefit provided by the separation between hw and sw, each mastered by specific companies with a deep knowledge of their own domain.
3. Based on your experience, what do you expect will be the predominant model for autonomous cars? Shared robotaxis or private vehicles? Or the two models will coexist?
Hard to predict because the future model will depend not only on technical capabilities and technological achievements but also on if/how the political system will decide to rule the field. Both scenarios are possible, and companies are now working on both models. The government can influence the deployment of one model or the other since it manages the public infrastructure (i.e., where vehicles move) and the possible incentives/taxes connected to both solutions. Consequently, the quickest governmental bodies of the most influential countries will shape the future of how this technology will be deployed.
4. Will the fierce competition in the sensors industry between visible, LIDAR, and Radar continue for much longer? Do you expect self-driving cars to continue using a combination of these technologies mainly, or will one of these sensor technologies emerge as the definitive winner?
Since currently there is no single and final solution to AD, including the definition of the sensing technologies, each car manufacturer, TIER1, or high-tech company is betting on their own specific mix of sensing technologies. Indeed, the more sensing is used, the denser the perception and the safer the final system. However, although this is a safety-critical system (i.e., safety is paramount), competition pushes for a careful evaluation of the technologies adopted.
If we look into the far future, when we have reached a sort of AGI (or at least common sense reasoning), we can mimic the human driver, who can handle the driving task using vision only. Until that moment, I guess we will need to seek the help of other technologies besides vision itself to compensate for humans’ lack of dense perception and intuition. Plus, some other technologies (for example, thermal sensing or perception in fog, rain, or very adverse weather) bring the additional advantage of surpassing human capabilities, which are limited to sensing in the visible and audible spectrum only. These technologies may one day complement the inputs to the AGI system to finally perform even better than the human driver.
5. Assume you must convince a skeptic to buy a self-driving car as soon as it becomes available commercially. What would be your strongest argument?
Definitely top-level safety, but increased comfort will also be a great buy for AD utilization.
6. In today’s autonomous car algorithms, what’s the balance between data from DBs and maps and information acquired in real-time from sensors? Has the balance between the two changed in recent years?
I’m familiar with this aspect since it is part of our history because we have been going through this transition in the last few years. Whenever we approach a complex problem, we usually try to find shortcuts. And Autonomous Driving is no exception: since the early years, a lot of effort was put into finding shortcuts, such as the use of very powerful -yet super-expensive- sensors or the adoption of extremely precise world maps that could help -and sometimes even replace- the sensing of the static world around the vehicle. Indeed, these shortcuts have paid off in the precompetitive and mainly research-based era; currently, in the consolidation phase, i.e., when the product’s cost and scalability are the main focus, those shortcuts need to be replaced by cost-effective and efficient alternatives.
In fact, this happened to our AD stack as well: a few years ago, our autonomous vehicles were pre-loaded with very precise HD maps, so that our environmental sensing was limited to detecting landmarks to allow for a precise localization on a map; driving in a static environment, then, was not based on real perception but rather replaced by navigating a pre-defined map. However, considerations like the availability of up-to-date and yet very precise maps for the whole world pushed us to leave high-definition maps behind in favor of a navigation system based on real-time perception. Creating a map while driving is definitely more complex than reading it from a database, but its undoubted scalability will pay off as a product.
7. Do you expect any significant spillovers of the technologies developed today for autonomous cars to other sectors?
Other important sectors, such as agricultural, industrial, mining and construction, and robotics in general, are indeed already benefiting from research done in the field of autonomous driving. Those other applications are, anyway, less sensitive to cost than automotive applications. For this reason, we already have successful applications of autonomous driving in those sectors. It goes without saying that any more cost-efficient technology, like better sensors or more powerful processing systems, that is being developed for automotive will also be deployed in other adjacent fields.
Conclusion
I would like to thank VisLab’s Alberto Broggi for sharing his profound knowledge and insights with me and my readers. His contributions to the field of autonomous vehicles are not only pioneering but also inspiring for the next generation of engineers and innovators. Alberto’s vision and relentless pursuit of excellence in autonomous driving technologies continue to push the boundaries of what we perceive as possible, paving the way for a safer, more efficient, and technologically advanced future. We thank Alberto for his invaluable participation in this interview and for providing a glimpse into the exciting advancements and challenges in the autonomous vehicle world.
Franco Folini Blog 
Leave a comment