This is an interview about SCALA 3 with Clement Nouvel of Valeo from July. I don't recall if it has been posted before, but it's worth a reread, but some extracts follow:
https://www.yolegroup.com/player-in...-ahead-an-interview-with-valeo-lidar-systems/
Compared to the previous generation, the Valeo SCALATM 3 optical head delivers an outstanding 50x points per second (more than any other player in the industry) and adds a 4th sensing dimension (passive image) synchronized in time and space* with the point cloud. This dual, very high-density output allows us to speak about LiDAR “images” and use Artificial Intelligence algorithms that were so far used for cameras, bringing LiDAR perception to an entirely new level, including low-level fusion.
Our software capabilities* allow us to provide an extensive suite of modules and LiDAR functions, from point cloud classification, object tracking, blockage detection, rain & spray detection, online calibration and misalignment detection, guaranteeing unparalleled availability, safety, and reliability.
...
Valeo reinvented the classical approach with the lessons learned from the past generations and the increased technological progress of the suppliers (Let’s not forget about ecosystem development, the only viable way to success!)**,
The interviewer asked about Valeo making their own chip, but CN pointed out the state of flux of the tech
. I
n addition, the LiDAR offering is evolving rapidly, quickly leaving products irrelevant to meeting customer needs.
The industry, from OEMs to Tier 2 suppliers, is still learning how best to use LiDAR on vehicles and how best to develop LiDAR components.
In such a market landscape, vertical integration might be tempting but often results in costly development costs, very risky ROI, and a heavy burden on a product roadmap that has to remain flexible.
Rather than betting on unpredictable volumes and standards, we are driving these changes today with the necessary agility that some components and solutions allow, especially coming from reliable, best-in-class automotive partners.**
Synergies across products, however, make sense for us as a system provider and a global leader in ADAS. In that sense, using common processing solutions between ADAS applications definitely brings a competitive advantage.***
Additionally, our products follow the larger Valeo strategy in terms of data processing centralization and the Software Defined Vehicle. For LiDAR perception, this means that most of the processing will likely migrate to a central computing unit***, making the LiDAR a satellite sensor. This will come with new business models and will facilitate a wider adoption of LiDAR on vehicles.
* TENNs?
** BrainChip since 2020
*** Nvidia/Qualcomm/????
So it looks like a Valeo lidar with integrated Akida SoC is not on the immediate horizon, as we had already suspected, but the references to SDV, ecosystem, software etc leave the door open for Akida 2/TENNs simulation software as we know Akida is processor agnostic.
Mercedes has also been developing its SDV, so again, as we suspected, maybe no Akida chips there for a while,
There are 6 levels of autonomous driving, and the industry is not yet up to level 4.
The adoption of EVs has slowed, so while EVs still need to optimize energy usage, maybe the efficiency of TENNs software will be utilized while the tech is evolving so rapidly. Hopefully we don't have to wait for Level 6 to see an SoC with Akida IP in ADAS.
There is a possibility that Nvidia or some other major processor producer could be persuaded to incorporate Akida2/TENNs IP in a future automotive processor. The other impetus for that is the massive power usage of cloud-based AI servers.
One other possibility is that the sensor makers incorporate AI in their sensors, as apparently some Chinese companies are doing. However, Valeo said they are going down the SDV route at present.
Another possibility is Luminar who have got forward orders from Mercedes. They have patents for the use of NNs with lidar, which, if implemented in software, would make heavy demands on the processor, but I don't recall that they have any patents for a NN SoC.
US12033334B2 Neural network for object detection and tracking 20200506
Then again, the processor load is a less important consideration in ICEs.
... so, of course, I had to search Luminar SDV, and it seems they use software AI:
https://www.luminartech.com/updates/luminar-launches-sentinel-software-for-automakers
Luminar launches Sentinel Software for automakers
SUNNYVALE, CA. – July 3, 2024 – Today Luminar (Nasdaq: LAZR), a leading global automotive technology company, announced it is launching Sentinel, its full-stack software suite, following five years of breakthrough developments in Software and AI systems. The Sentinel solution enables automakers to accelerate development of advanced safety and autonomous functionality on vehicles, and includes Proactive Safety,
Perception, 3D Mapping and Localization, Simulation, and
Dynamic LiDAR features. The
first shipments occurred this week, and are planned to be delivered to top automakers in the third quarter of this year.
Luminar’s Sentinel suite is a solution currently in-development to enable advanced safety and autonomous functionality with automakers,
leveraging Luminar’s LiDAR hardware and AI-based software technologies developed primarily in-house by its software teams in Silicon Valley and Munich, Germany, with support from 3rd party software partners including Scale AI, Applied Intuition, and Civil Maps which Luminar acquired in 2022. Sentinel differs from other full-stack autonomous software solutions from other companies that develop software for driverless robo-taxis. Instead, Luminar’s development has focused from the beginning on giving automakers the toolkit of capabilities outlined above to implement next generation safety and autonomous capabilities for consumer production vehicles. With the launch, Sentinel is targeted to be assessed by and integrated into automaker software development efforts.
...
“People know Luminar for pioneering LiDAR systems for global production vehicles, but s
ince 2017 we have also been building next-generation AI-based software for improved safety and autonomy,” said Austin Russell, Founder and CEO of Luminar. “The majority of major automakers don’t currently have a software solution for next-generation assisted and autonomous driving systems. Our launch couldn’t be more timely with the new NHTSA mandate for next-generation safety in all US production vehicles by 2029, and as of today, we’re the only solution we know of that meets all of these requirements. This gives us the opportunity to dramatically expand our content value per vehicle in a high-margin capacity.”
...
Sentinel was first previewed by Luminar as a prototype in 2021 with a fraction of its current capabilities. The
Sentinel software included in the kits is in continuous development, with the intent to be evaluated by automakers and technology companies for potential integration into future advanced safety and autonomous driving programs. Sentinel is built to be compatible with almost any consumer and commercial vehicle, including
both combustion engine and electric drive train models.
...
1
Sentinel has not reached technological feasibility and remains in development. The Development Kits are not working models of a production release version.
”?
