The automotive sector has mostly stopped blowing its own horn, ending the heralding of the latest advancements that it had made in self-driving or connectivity technologies. To this end, it’s been quite quiet in the past few months, in terms of news announcements, from both vendors and customers. However, Nvidia’s annual conference was the stage for a raft of such news, as the company tries to get over a crash in its share price in the fallout from the bitcoin crash.
Nvidia announced a new collaboration with the Toyota Research Institute – Advanced Development (TRI-AD), which will see the pair work on developing and validating self-driving vehicles. This new deal builds on TRI-AD’s parent company, Toyota, using Nvidia’s Drive AGX Xavier, an autonomous vehicle (AV) computer that Nvidia envisions being the heart of future AV designs.
Toyota is far from the only such hardware customer for Nvidia, but it does look to be one of the most involved customers. Together, the two are aiming to develop a computing design architecture that can be ‘scaled across many vehicle models and types.’ This platform would feature Nvidia’s silicon, include both the AGX Xavier and AGX Pegasus, as well as the suite of supporting software that Nvidia slings with its hardware.
But Toyota and TRI-AD are going to be the first customers for Nvidia’s new Drive Constellation offering, which has just been made generally available. The cloud-based system is a GPU-powered array that facilitates simulation, which lets automakers like Toyota test how their AV designs will operate in the real-world. Simulation lets users create huge virtual fleets of test cars, which are very valuable in machine-learning applications, as you can ‘gather’ far more driving data with the simulations than you could with a real-world test fleet – running through video and sensor data at much faster speeds than you could out there in the real-world. Now, Nvidia is offering AV simulation as a service, via the new Constellation suite.
“Our vision is to enable self-driving vehicles with the ultimate goal of reducing fatalities to zero, enabling smoother transportation, and providing mobility for all,” said D. James Kuffner, CEO of TRI-AD. “Our technology collaboration with NVIDIA is important to realizing this vision. We believe large-scale simulation tools for software validation and testing are critical for automated driving systems.”
As for the more technical details, Nvidia says that Constellation comprises two joint servers – the Constellation Simulator, and the Constellation Vehicle. The former is essentially creating the associated sensor outputs that a real-world car would create while driving, using GPUs and the Drive Sim software, and then feeding this data stream to the latter, which houses the AGX Pegasus AV computers that would be used in the real-world car. The outcome from the virtual vehicle is then fed back into the simulator, to create a complete loop. This then lets the user see how the Constellation Vehicle would react if it were actually out on the road.
We understand that Waymo has a pretty similar set up, but of course, Waymo isn’t making that available to other players in the AV space. Nvidia’s desire to sell more silicon necessitates the expansion of its supporting services, such as Constellation. It needs to ensure that customers can most easily adopt its designs in their vehicles, and Nvidia isn’t fussy about whether it works directly with automakers or with the OEMs that supply them.
TÜV SÜD was namedropped in the announcement. The safety agency is using Constellation as a way to design and test the self-driving validation standards it will use. “TÜV SÜD is looking for simulation tools that are trustworthy, robust and scalable for the approval of autonomous vehicles,” said Houssem Abdellatif, global head of Autonomous Driving and ADAS at TÜV SÜD. “NVIDIA DRIVE Constellation provides a powerful and highly scalable solution to achieve this goal.”
The final bit of automotive news that crept out of Nvidia’s GPU Technology Conference (GTC) was a new set of driving algorithms for AVs, which will be added to the Drive AV suite. The Safety Force Field (SFF) is a new policy that aims to enable more defensive driving, to protect the vehicle and other road users. As Nvidia puts it, the SFF framework ensures that its actions will never create, escalate, or contribute to an unsafe situation.
It says it has tested SFF using recreations of incidents that can’t be safely done on public roads, such as collisions and near-misses. The design prioritizes ‘one core principle of collision avoidance, as opposed to a large set of rules and expectations,’ which sounds like Nvidia has tried to simplify the web of decision-making rules used in previous frameworks. Nvidia says that SFF is an open platform that can be combined with any other driving software.
“By removing human error from the driving equation, we can prevent the vast majority of collisions and minimize the impact of those that do occur,” said David Nister, VP of Autonomous Driving Software at Nvidia. “SFF is mathematically designed such that autonomous vehicles equipped with SFF will, like magnets that repel each other, keep themselves out of harm’s way and not contribute to unsafe situations.”