The Self-Driving Startup Nobody Saw Coming | E2289

Introduction to Wayve and Its Self-Driving Technology

• The company Wayve, a UK-based self-driving startup, has been working on end-to-end learning for self-driving cars, an approach that was initially widely dismissed, but is now gaining industry support, with investments from companies like Nvidia, Qualcomm, and Uber 10s.
• Wayve has partnered with multiple manufacturers, with a minimum of 25,000 partnerships, and its volume is comparable to double the number of cars Tesla builds in a year, which could significantly impact the demand for cars without self-driving capabilities 42s.
• The company's technology enables every car to be intelligently driven by a machine that never blinks, essentially providing a private chauffeur for each car owner, and Wayve is not for sale to any other company 1m9s.
• Wayve's world model is a powerful representation learning method that allows the company to learn a representation of the world that cares about what matters, such as road lines, curbs, and traffic signals, and can be used as a simulator to validate or control what's in front of the car 2m6s.

World Models and Simulation in Self-Driving Technology

• Wayve's world model is a powerful representation learning method that allows the company to learn a representation of the world that cares about what matters, such as road lines, curbs, and traffic signals, and can be used as a simulator to validate or control what's in front of the car 2m6s.
• The company has released new world models, including Gaia 2 and 3, which have improved upon the previous models, and the CEO, Alex Kendall, believes that the industry is now getting behind Wayve's contrarian approach to self-driving cars 4m21s.
• Wayve's approach prioritizes generalization across many environments over driverless optimization in a single domain, and the company has been working on this approach since 2017, with its first blog post discussing a world model with 20,000 parameters 5m15s.
• The self-driving technology utilizes a virtual world with obstacles, traffic, weather, locations, and rules to test and improve the driver's abilities, allowing for infinite miles of testing in a virtual setting, which would be time-consuming and costly in the real world, with significant safety implications 10s.
• The world model used in self-driving technology is similar to the human brain's ability to replay experiences and learn from them, but it is a more complex and rich representation of the world, and one of its best uses is as a simulator for testing and improving the driver's abilities 42s.
• The development of self-driving technology is an arms race between learning a driving policy and learning a simulator, and having one means solving the other, with end-to-end learning being the best approach for both learning driving policies and simulating, as it allows for the use of data to model complex and diverse scenes 2m6s.
• The world model learns from hundreds of petabytes of data, including internet-scale data, dashcam footage, and data from over a dozen automakers, and it has been improved algorithmically to understand multiple sensors, including cameras, radar, and lidar, and is controllable, allowing for the simulation of real-world scenarios and adversarial testing 2m6s.
• The AI driver can work with various types of sensors and information, including lidar, radar, and visual data, and there is no minimum level of ingestion required, as the goal is to be the intelligence layer across any vehicle, anywhere, with the ability to take in any type of information and make decisions based on it 2m6s.

Sensor Integration and System Capabilities

• The approach to self-driving supports various sensor modalities, including camera-only, radar, and lidar, and can learn to understand the capabilities and limitations of different sensor architectures through a world model 10s.
• There is a minimum bar of safety required for hands-off, eyes-off, or driverless systems, and while it is possible to achieve all levels with a camera-only system, it may be faster and more efficient to use a combination of sensors, such as camera, radar, and lidar 2m6s.
• Many partners building robot taxis use a combination of camera, radar, and lidar, but these are automotive-grade, mass-market, low-cost sensing devices, rather than custom or bespoke solutions 4m30s.

Funding, Partnerships, and Market Expansion

• The company has raised a significant amount of money, $1.2 to $1.5 billion, and has made significant technological progress, with two new world models released, and is now entering an integration and product deployment phase 6m40s.
• The company's mission is to bring intelligence to any vehicle anywhere, and with the newly raised capital, they plan to deploy their technology in supervised robo-taxi trials in multiple cities, including London and Tokyo, and in consumer vehicles starting next year, with support from partners like Nissan, Mercedes, and Stellantis 8m50s.
• The company's technology will be integrated into Nissan's consumer vehicle lineup, as announced last year, and they are also working with other partners to bring their technology to market 10m40s.
• The announcement of the robo-taxi earlier this year was made because automakers want to work with partners across L2, L3, and L4, which helps speed and efficiency and allows for data and integration leverage 10s.
• Nissan announced that they will bring the technology to 90% of their vehicles, which is approximately 2.7 million cars per year, a volume that is double the number of cars Tesla builds in a year 2m6s.
• There are three ways to bring autonomy to market: building your own cars, building your own fleet city by city, or licensing the technology to any fleet or automaker, with the latter being the chosen business model due to its potential for large volume 4m6s.

Business Model and Market Strategy

• The company has built a flexible and generalizable AI driver that enables a different business model, which is licensing the technology to any fleet or automaker, and this approach is possible due to the technical strategy and business model 5m30s.
• The question of whether the company has cracked self-driving was answered by stating that while technological progress has been made, the economics of bringing this to the world have not been sorted out yet, and there are still technical and economic questions to be addressed 6m40s.
• The company has made progress in generalized self-driving, and while there is still some technical risk, the main challenges now are engineering execution risk, product integration risk, and deployment risk, particularly in integrating the technology into vehicles with the right infrastructure 10m30s.
• The company has shown that it can achieve hands-off driving and has demonstrated the level of performance needed for a delightful product, with Tezza announcing $1.5 billion in revenue per year, but there is still a gap in performance to achieve general-purpose driverless technology 12m10s.
• The path to achieving general-purpose driverless technology is clear, and the company is focused on integrating the technology into vehicles and addressing the engineering execution and product integration risks ahead 14m40s.
• Programs are underway with major manufacturers to develop self-driving technology, and the next steps involve scaling up the AI model to achieve the required level of performance, which is a predictable process that requires data, compute, and algorithmic innovation 10s.
• To validate the technology, engineering activities need to be scaled up to prove that it is safer than a human driver, and regulatory sign-off is required to launch the products, with regulators in some countries and states already putting regulations in place ahead of the products being ready 42s.
• The UN has legalized a legal pathway for L3 and L4 driving, covering most countries except the US and China, providing a pathway for deployment, and the technical challenges to get from hands-off to eyes-off driving are solvable with data, compute, and algorithmic innovation 2m6s.
• The business model for self-driving technology involves working with manufacturers and demand providers, and the cost to the consumer is still being figured out, with possible options including a one-time fee to the manufacturer or a regular payment for ongoing improvements and data exchange 4m30s.

Pricing Models and Consumer Adoption

• Different models are being explored, including bundling the technology with the car and including it for free, or passing through economics to the technology provider, with some manufacturers looking to include it as a standard feature like a seatbelt 6m30s.
• The automotive industry is exploring different pricing models for self-driving features, including one-time fees, recurring subscriptions, and free trials, with Tesla charging $100 a month for its features, and the industry is likely to move towards a subscription model due to ongoing over-the-air updates and insurance costs 10s.
• The subscription model will allow car owners to pay for their own private chauffeur, with the intelligence running on the edge of the car, and manufacturers will bear ongoing insurance costs for L3 or L4 driving 42s.

Liability, Insurance, and Legal Frameworks

• AI tools are making it easier for solo founders to run their own businesses, and having a beautiful and attention-grabbing website is crucial, with Squarespace being a recommended platform for designing and launching a business 2m6s.
• Insurance costs and liability for self-driving cars are still unsettled, with factors such as the level of autonomy, regulatory environment, and commercial contracting determining who holds the liability 4m30s.
• In the case of a hands-off system, the driver remains liable, while for eyes-off or driverless systems, the manufacturer or operator holds the liability, with some insured and contracted liability flowing through to various parts of the ecosystem 6m15s.
• The company is not planning to build an insurance product, but rather focus on its core offering, with insurance companies like Chubb or Berkshire Hathaway potentially handling the insurance element 8m30s.

Market Potential and Industry Growth

• The market for self-driving cars is enormous, with around 100 million vehicles produced each year, and it is estimated that a significant percentage of newly manufactured cars will be made with the capacity to work with self-driving products like Wave in the next 5 years 10m45s.
• The current penetration of advanced self-driving technology in the market is around 15%, but most of this is limited to basic systems like highway lane keep assist, with only a small fraction of the market, primarily Tesla, offering a full self-driving experience 10s.
• Luxury manufacturers are starting to incorporate high-level compute systems, such as those from Nvidia or Qualcomm, into their vehicles, and volume manufacturers like Nissan are planning to introduce similar technology in their vehicles from 2027, which will significantly improve the driving experience over time 42s.
• It is expected that within 5 years, a significant portion of the market will have vehicles with advanced self-driving hardware, and the technology will continue to improve, eventually becoming a standard feature in all vehicles, with even the most basic cars likely to have some level of autonomy due to regulatory requirements and road safety concerns 1m6s.

Financial Position and Strategic Partnerships

• The potential market for self-driving technology is enormous, with 100 million new cars sold every year, and even a relatively low monthly subscription fee of $100 could generate $100 billion in revenue annually 2m6s.
• Waymo's work on both robotaxis and consumer cars provides an advantage, as the data from consumer cars can be used to develop general-purpose robotaxis, and the company's relationships with OEMs can help to create a high-margin software business with global supply chain and geography scale 3m6s.
• Waymo is in a strong financial position, with over $2 billion in capital, which should provide enough runway to get into production with an OEM and achieve early volume without needing additional funding 4m6s.
• The company has a strong set of shareholders and sufficient capital to deploy its business and achieve a free cash flow positive scale velocity, with decade-long relationships established with automakers, providing confidence in the company's security and potential to reach escape velocity without needing further raises, at least for now 10s.

Scalability and Expansion into Other Domains

• The company's technology is considered scalable and can be applied to various robotics applications, including self-driving trucks, construction equipment, and other earthmovers, with the potential to learn behaviors in these domains using a small amount of data and the company's foundation model 42s.
• The company started with the hardest application, consumer vehicles, to build the most scalable technology, and has done proof of concepts in areas like sidewalk delivery, trucking, mining, and warehouse logistics, with its simulator, Gaia, adaptable to these domains 2m6s.
• The company has spent the last 5 years building on its end-to-end learning demo, focusing on making the technology safety-qualified and compliant for the automotive industry, to ensure it is safe, validatable, and runnable in an embedded environment 6m34s.
• The company's technology stack, including the driving policy, reinforcement learning, and simulation stack, can transfer to other domains with a small amount of data, allowing for potential expansion into new areas 4m30s.
• The company's approach has allowed it to tackle the hardest problem first, building a scalable technology that can work with any robotics application, and has established a strong foundation for future growth and development 3m15s.

Company Growth and Hiring

• The company Wave is working on self-driving technology and plans to become the intelligence layer across every robotic vertical, with automotive being the initial focus, and this expertise is expected to scale nicely due to the challenge of working in major automotive centers 10s.
• Having world models, simulation experience, and access to data can enable the automation of various tasks with wheels, and the limit to this is largely a question of data and investing time to bring it to market 42s.
• Mobility is expected to become more prominent than manipulation robotics, as there is already a established tech stack and platforms in automotive, with millions of cars being built, and this can be scaled to other areas such as warehouses or airports 1m6s.
• Manipulation robotics will likely come second to mobility, as it requires platforms at scale and data, but it is expected to adapt quickly with the data obtained from mobility 2m6s.
• Wave is hiring and growing, with high demand from the automotive sector, and the company's unique technology at the intersection of frontier embodied AI and automotive is attracting interest from every car manufacturer 8m10s.
• The company can be found on all social platforms by searching for "Wave", and people will be able to experience their technology through fleets in London, Tokyo, Stuttgart, and the Bay Area, or by buying one of their cars from next year with partners like Nissan 9m20s.

World Models and Simulation in Physical AI

• The company Waabi has built an environment that brings together two cultures, typically considered incompatible, to work on frontier AI and automotive production-grade technology, allowing employees to see their work deployed in consumer products at a large scale 10s.
• Waabi offers various roles across the full stack, including machine learning, data, software, product, application, validation, operations, public policy, and other enabling functions to unlock the future of self-driving technology 42s.
• The company has raised $2 billion in funding and is looking for people to join and help spend this investment to bring self-driving technology to both trucking and cars 1m30s.
• Waabi founder and CEO, Raquel Urtasun, discusses the importance of world models in self-driving technology, which involves building representations of the world to enable simulation systems that are as realistic as the real world 2m6s.
• World models are not just about creating interactive worlds, but also about having controllability over what they generate, which is a key differentiation in building world models for physical AI versus other applications like video games or movies 4m30s.
• The AI model in self-driving technology is put into virtual situations, reacting to the environment created for them, which may seem similar to a video game from the AI model's perspective, but is actually a critical component of training and testing self-driving systems 6m40s.
• The development of self-driving vehicles relies on creating world models that accurately represent the physical world, allowing the vehicle to interact with its environment and make decisions, which is crucial for the vehicle's safety and reliability 10s.
• World models enable the testing and training of autonomous systems at scale, in parallel, and in the cloud, reducing the need for years of experimentation in the real world and increasing the safety of the system 1m42s.
• The use of world models can reduce the time to market for self-driving vehicles, increase their safety, and cut down on the costs associated with integrating thousands of engineers over time, regardless of whether high-definition maps or different sensors are used 2m6s.

High-Definition Maps and Sensor Integration

• The debate between using high-definition maps and other approaches is not about whether to use them, but rather about how to build them in a scalable and efficient manner, with the help of AI, to provide an additional layer of safety 4m10s.
• The use of high-definition maps is not mutually exclusive with other sensors, and self-driving vehicles can react and drive safely even if the maps are wrong or outdated, making it a worthwhile investment for increased safety 5m40s.
• The cost of building and using high-definition maps, including the bill of materials (BOM), is an important consideration, but with efficient and scalable technologies, the benefits of using them outweigh the costs 7m20s.

Expansion into Different Use Cases

• The company started with self-driving trucks on highways, then expanded into surface streets, and with their latest series E announcement and the Uber deal, they are moving into robo taxis, and the original world model foundation of the company has made it easier to expand into different areas of automation 10s.
• The physical AI platform, which includes the world model, simulator, and autonomy system, was built from the beginning to be utilized for multiple physical AI use cases, allowing the company to capture multi-trillion dollar markets 2m6s.
• The autonomy system is verifiable end-to-end technology, which is different from traditional AV 1.0 or 2.0, and it has been a massive accelerator for the company, enabling them to use the same brain and simulator for different use cases, such as robot taxis and trucks 2m6s.
• The technology allows for an additive approach, where each program accelerates the other, and it enables the company to handle different driving needs with a single intelligent mind, similar to how humans use one brain for all their driving needs 4m30s.
• While human brains are not very specialized, the company's single brain approach can handle different use cases, such as trucking and driving cars in the city, without needing to be super specialized in terms of technology 6m20s.
• The company believes that specialization may make sense for other types of skills that are more different, but for perceiving and understanding the world in 4D, reasoning, and action, the core characteristics are common to everything and can be handled by the same brain 8m40s.

Industry Trends and Technological Advancements

• The self-driving industry has experienced a significant explosion of capability in the last two or three years, especially in the last year, similar to the recent advancements in AI, with new products, features, and capabilities emerging 10m50s.
• The development of self-driving technology has been impacted by fundamental changes in AI models and intelligence, with three key factors converging to enable scalable and safe deployment, including hardware and OEMs, regulatory frameworks, and consumer acceptance, which is evident in the success of Waymo deployments 10s.
• The regulatory frameworks are evolving to enable the deployment of self-driving technology, and consumers are embracing it, as seen in the adoption of Waymo, with people understanding that the technology makes roads safer and is a better product than human driving 2m6s.
• The trucking industry is a prime example of where self-driving technology can have a significant impact, with a clear case for adoption due to driver shortages, safety issues, and cost savings, making it a no-brainer for companies to invest in this technology 4m42s.
• The advancements in AI technology are also driving the acceleration of self-driving development, with next-generation companies having a second-mover advantage, and the ability to build more powerful and generalizable AI capabilities, enabling faster expansion and solving long-tail problems 6m15s.
• The combination of market preparedness, demand, regulatory structures, consumer uptake, market fit, and AI improvements are all coming together to make this an exciting time for self-driving technology, which is expected to change the way the world works, with transportation being at the center of everything 8m30s.

Funding and Strategic Vision

• The company's recent Series C funding round raised over a billion dollars, which may seem excessive given its asset-light and efficient nature, but this investment provides stability and allows the company to make long-term bets and investments without compromising its goals 10s.
• The funding enables the company to think about the future of the market, make strategic investments, and be robust to any potential delays or ecosystem changes, while also exploring additional opportunities such as robot taxis 2m6s.
• The company's strategy has been focused on building for scale from day one, investing heavily in foundational technology, which has placed it in a strong position for widespread adoption of its technology 4m30s.
• The company will continue to focus on the autonomy layer and partner with other manufacturers for car and truck production, rather than becoming an original equipment manufacturer (OEM) itself, as it believes this is the safest and most effective path to market 6m40s.
• The company's stability and long-term thinking allow it to prioritize safe self-driving technology and resist pressure to compromise on short-term progress, instead focusing on building a strong foundation for the future 8m50s.

Product Development and Market Entry

• The focus of Waze was on building a product that solves pain points and addresses customer needs, rather than starting with a large number of operations and commercial operations 10s.
• The industry initially adopted a hub-to-hub model, where autonomous trucks drive between hubs and humans complete the trip, due to the difficulty of driving on surface streets, but this approach resulted in a product that customers did not want 1m20s.
• Waze invested in next-generation AI technology to enable trucks to drive on surface streets, allowing them to provide a better product that can go to the end customer's door 2m30s.
• Wabi is currently in the commercialization phase of its self-driving technology in the trucking space, with multiple trucks on the roads and a massive partnership with Uber Freight for billions of miles of deployment 4m10s.
• The company has a decent-sized fleet of self-driving vehicles, with double-digit numbers of trucks, and is working with its OEM partner Volvo to bring its technology to market 5m20s.
• Volvo has stated that it is quarters away from deploying Wabi's fully redundant and validated platform, with hundreds of trucks expected to be deployed by 2027 6m40s.

Business Models and Revenue Streams

• There are two potential ways for Wabi to charge for its technology: selling the system to its OEM partner or offering it as a service, and the company is considering the compute-intensive requirements of operating its AI drivers 8m10s.
• Woven's technology, including its world model and autonomous system, is highly efficient, which can be seen in its advanced technology and upcoming driverless launch with the IM, and this efficiency is a key aspect of the company's business model 10s.
• The company's focus is on providing sustainable and efficient solutions through next-generation technology, which is at the core of its DNA, and it has been innovating in this field for over two decades 2m6s.
• Woven operates on a "driver as a service" model, providing technology for both trucking and robotaxi services, and it does not plan to own and operate trucks or robotaxis, instead partnering with companies like Uber, which plays a fundamental role in the go-to-market strategy for robotaxis 2m6s.
• The company's business model is based on a recurring fee, with payments made on a per-mile basis, which incentivizes all parties involved to work together and use the technology efficiently 4m42s.

Partnerships and Go-to-Market Strategy

• Woven has announced a partnership with Uber to provide technology for a minimum of 25,000 robotaxis, and while the company has not announced an OEM partner for this deal, it has stated that its partnership with Uber is a key part of its go-to-market strategy 8m10s.
• The company's partnerships, including those with Uber and Volvo, are important for its business model, and it is able to work with a variety of partners, including OEMs and companies like Walmart, to provide its technology and services 6m40s.
• The company has a partnership with Uber, where Uber plays the market component, and there is an OEM that will provide a redundant platform for vertical integration, which is believed to be the safe and scalable path to market 10s.
• The OEM has not been announced yet, but there are a few OEMs that have a platform ready, and the ecosystem is excited about partnering with the company 1m4s.
• The company is excited about its entry into robot taxis and is looking forward to sharing more details about its partnership with the OEM 2m6s.
• The company's founder worked for Uber's ATG for 4 years, and the company has a partnership with Uber Freight and Uber's taxi service for robot taxis, but the company is not for sale 4m14s.

Founder Vision and Future Goals

• The founder's goal is to build a physically empowered house that is transforming the world, and the company is focused on building a world model approach to solving autonomy 6m30s.
• The company does not think that level 4 or level 5 self-driving cars will be available for personal purchase in the next 3 years, but robot taxis at scale may be available 9m40s.
• The company believes that the traditional approach to developing self-driving technology, where level 2 comes before level 3, and level 3 comes before level 4, may not be the fastest or most effective path 11m40s.

Level 4 Autonomy and Deployment Plans

• The company has developed a level four native technology, which is a gigantic difference from a level two plus product, and this difference is crucial to understand because it solves a totally different safety problem 10s.
• The level four system requires no human intervention, and the company has built a technology that can achieve this, which is not just about driving well or having few interventions, but rather a new set of metrics that matter for level four 42s.
• The company is planning to deploy more than 25,000 robotaxis onto the market and streets, and they will announce their future plans and timeline soon 2m6s.

Company Resources and External Opportunities

• The company's website is where people can go to learn more about their technology and innovations in physical AI, and they are massively expanding and looking for partners 2m6s.
• There are also various resources available for startup founders, including Founder University Cohort 13, the launch accelerator, Jason's Angel Syndicate, and This Week in AI, which provide guidance, investment, and access to quality deal flow 2m6s.