Interview with Glen DeVos, CEO of MicroVision
“Lidar sensors are too expensive for mass adoption”
Glen DeVos, CEO of MicroVision, sees the breakthrough of lidar less as a question of performance and above all as a cost and integration task. With more than four decades of experience in automotive, mobility, and vehicle electronics, he is driving MicroVision's path from development to scaled commercialization.
MicroVision
SDVs, AI and LiDAR are set to transform the car. However, the path to scalable platforms is still full of challenges. In this interview, MicroVision CEO Glen DeVos explains where the architecture, value creation and lidar technologies still fall short.
As the complexity of software-defined vehicles increases, the challenges facing the automotive industry are shifting. It is no longer simply about individual control units, sensors or software functions, but about scalable E/E architectures, clear make-or-buy decisions and robust validation methods for AI-based, safety-critical systems. At the same time, the question arises as to which technologies will actually make the leap from development logic to broad industrial deployment.
This is precisely where Glen DeVos, CEO of MicroVision, comes in with his presentation “Lidar 2.0” in the “Technologies to Watch” category at the 30th AUTOMOBIL-ELEKTRONIK Kongress. In the run-up to the event, we spoke with him about why software-defined vehicles require a fundamental realignment of vehicle architecture, how software decisions will shape value creation in the automotive ecosystem in the future – and why, although lidar is technically convincing, its breakthrough into series production will be decided elsewhere.
Looking ahead three to five years, what will be the
biggest bottleneck in turning SDV and AI strategies into scalable,
industrialized vehicle platforms?
In my view, there are three fundamental challenges that need
to be addressed or resolved in the coming years in order to turning into
scalable vehicle platforms. First: Ability of the OEM’s to implement the underlying
vehicle electrical and compute architecture in a cost effective and scalable
manner – SDV’s require a complete change in how compute, power distribution,
communication and deterministic actuation are managed – this is a massive
overhaul and a significant challenge for the OEM’s to manage and you can’t
iterate your way there. Second: To create a scalable vehicle platform, it is essential to
establish the overall definition of the software architecture and then the
OEM’s chose, which development tasks will be handled in-house, and which will
be outsourced? What should the overall development environment and toolchains
look like? Third: Finally – the complexity of validating the SDV performance, for
embedded systems where physical AI is now being used for deterministic and
safety critical functionality, like ADAS or autonomy. Physical validation is
simply not practical, so virtual testing and validation are key enablers but
are still nascent in their usage.
Which decision being made today will most strongly
determine where value is created in the future automotive ecosystem?
The decisions that are made today with the greatest impact
to value creation relate to software content development – specifically, what
do the OEM’s “insource” and what do they rely on partners or suppliers to
provide. While hardware, in general, continues to follow historic trends of the
manufacturing value stream stratification - from raw materials to components to
sub-systems, to the vehicle assembly -, software is very different and has
disrupted the traditional value creation hierarchy. As a result, software
content providers now have the opportunity to create significantly more value
with the vehicle – in particular the software defined vehicle – and the OEM’s
decisions around what they “make vs. buy” has a major impact on the value
stream.
Where do current approaches to SDVs and
next-generation E/E architectures still fall short in real-world programs?
There are several areas where SDV’s and NG architectures
fall short as follows: First and foremost is carrying forward legacy
architectural elements, networking requirements/protocols, and development
tools that limit and/or inhibit SDV architectures. Furthermore, there is a lack
of development of the off vehicle ecosystem. If you think back: how long has it
taken for OTA to really become standard for the mass market? The third weakness
are the OEM procurement systems that still operate as if we are sourcing hardware
components. The approaches are still very inconsistent and ineffective,
particularly when it comes to procuring and managing software content on the
vehicle.
How is lidar technology evolving to support
next-gen ADAS systems?
First and foremost – lower cost. Current lidar sensors are
simply far too expensive for mass adoption. Unless costs are significantly
reduced, lidar sensors will not become a standard part of vehicle perception
systems.
What are the key challenges in scaling lidar for
production vehicles?
It is not the costs that follow volume, but rather volume
follows costs. The lidar performance is not the issue, rather it’s the cost of
the sensor. Lidar sensor costs have to be reduced to a level that allows
OEM’s toadd them to the vehicle perception system and create value for the end
customer, unlocking user features that the end consumer must have. In the case
of ADAS this was for example Adaptive Cruise Control, Blind Spot Detection,
Bird’s Eye View, Park Aid - along with regulatory requirements. For L2+/L3, where
Lidar can be an enabler, these features simply haven’t moved the market. So
far, the end consumer hasn’t seen sufficient value in the vehicle to justify
the added cost.
Where does lidar still fall short compared to
expectations?
Today’s automotive lidar are too big, too power hungry, and
to expensive. They work very well and their range, resolution, etc. can do the
job. But they are simply too expensive and difficult to integrate into the
vehicle and system architecture.
