There is certainly a lot of activity and interest in the AI and autonomous driving sectors recently! And that is certainly true here at Motional, although some of you may be wondering, “where has Motional been?” As AI has rapidly evolved, so have we, and I couldn’t be more proud of the incredible work our team has done to position us for the future by adopting and integrating the latest breakthroughs into our AI-first autonomous technology stack, which is already being tested on the road.
It’s a privilege to now lead the Motional team, build on an impressive history of achievement in autonomous vehicle development and strengthen our strategic advantages.
To bring you back up to speed on where we are in our journey:
- We achieved our first driverless milestone in 2023 with the Hyundai IONIQ 5 robotaxi in Las Vegas and have continued to expand our capabilities ever since.
- We are currently operating a range of autonomous testing in Las Vegas, driverless and with safety operators as we continue to advance our embodied AI capabilities and grow our operational area into new parts of the city.
- Hyundai remains a strong and committed partner demonstrated by development and manufacturing of the IONIQ 5 robotaxi as an FMVSS-certified fully driverless vehicle being built on an assembly line starting back in 2023.
- We have conducted a total of more than 130,000 customer rides through our ride hail partnerships starting in 2018 and expanding in 2022.
- In advance of driverless operations, Motional earned a first-of-its-kind endorsement in 2021 to operate without a driver on public roads, because safety should be paramount to any autonomous vehicle development effort.
This list is only a fraction of our achievements, but it reinforces the strength and experience of this team, and reflects elements of what will propel us into this new chapter of the autonomous vehicle industry.
So, what have we been up to? With the breakthroughs in embodied AI, and foundation models, it became clear there were gaps in the technical approaches of those of us that initially formed the industry. Even though most of the AV tech already on the road was enabled by deep learning, it wasn't capable of scaling up in a cost effective way. We now saw new possibilities for achieving affordable and scalable driverless autonomy.
If you look at AV technology architectures in the field today, many are using specialized machine learning (ML) models that are impressive in operation, but are very complex, thus expensive and slow to update in support of expansion to a large set of cities. On the other hand, fully end-to-end AV technology solutions significantly reduce the amount of models, but they don't give engineers enough introspection into what's happening in order to truly achieve expected safety standards across important edge cases. The general end-to-end only approach can get to a really good 80-90% - maybe even 95% - solution but that’s not good enough to remove a driver or to earn the trust of cities, communities and customers.
Instead of rushing to end-to-end AI solutions, we're taking a different path. Motional's time in the industry resulted in having the deep expertise needed to translate recent AI breakthroughs into safe autonomous driving solutions, so we will be able to build to scale with lower cost and less dedicated development for new city deployments.
We've now combined the best of state-of-the-art innovations in AI with tried and true safety backstops to create a comprehensive technical solution for responsible and affordable AVs.
Autonomous vehicle technology must evolve to leverage embodied foundation models - a very powerful, adaptable “brain” - versus individual machine learning (ML) models for specific tasks, because that’s how to achieve generalizable behavior more naturally and adapt to many tasks without costly development or inefficient training. For instance, developing a small model for a specialized task can be done rather quickly, but maintaining it over time is where it can get complicated. When entering a new city or neighborhood, each model will need to go through its own lifecycle: data curation, model development and evaluation plus analysis and fixing of corner cases. So, this approach can result in the creation of teams and resources dedicated to each specialized model, and that can lead to organizational complexity and unnecessary costs.
Motional is developing a set of LDMs (Large Driving Models) to introduce embodied foundation models into our autonomous vehicle technology stack. They are designed to be trained efficiently and are capable of optimized learning across vast datasets resulting in being prepared to safely perform driverless operations in new environments.
Our LDMs include the right number of foundation models to achieve driverless safety benchmarks with efficiencies that reduce the costs needed to quickly scale across regions and across the globe. Our data set is already extremely diverse based on our current testing operations in Las Vegas and Pittsburgh, plus our experience from testing in Boston, Los Angeles and Singapore.
Because we re-architected to leverage LDMs and optimized the amount of machine learning models in our AV tech stack with the right learning capacity, we also have enough introspection to really understand what's happening so that we can more easily improve the system and solve long tail issues. Instead of just adding another ML model as we identify new long tail issues, we can understand where in the stack the issues are stemming from and intelligently tailor our distribution of edge cases and model design to achieve the critical performance improvements needed to safely deploy a SAE Level 4-capable automated driving system at scale.
While we may be entering a new phase, some things remain the same for each of us trying to grow this industry, and that’s putting safety above everything else. Those of us developing autonomous technology have a collective responsibility to foster this industry, from technical advancement to regulatory coordination to consumer confidence. We need to maintain an overwatch of safety outcomes, since that is core to earning trust. Trust takes time to build, and the commercialization of autonomous vehicles is a long game. There’s no easy way around it.
Hopefully you now feel more up to speed on what’s going on at Motional, and have a better idea where we’re headed - on a clear path to build cost-efficient, adaptive and safe autonomous vehicles. And to get there, we’re growing, so we’d love to have you join us. Learn more about open positions here.
There’s much more to show and tell about our plans and I look forward to sharing more soon.
