Physical AI has become the topic du jour, with capital pouring into robotics, humanoids, and automation. At G2, we believe vehicle autonomy is the most tangible — and arguably most underrated — manifestation of physical AI happening today. The progress made over the last decade has been nothing short of inspiring considering not too long ago many had dismissed the problem as impossible or decades away. As of today, Waymo is run-rating at >450K paid rides per week across six cities, Tesla has accumulated >6B FSD miles, and Zoox is offering rides in its purpose-built robotaxi on public roads. While the road has been bumpy, we should take a moment to reflect on how far the industry has come.
The debate is no longer whether autonomy will work, but how it can scale safely, efficiently, and cost-effectively. And with early commercial success in robotaxi, the industry can turn its attention to the next frontier: how we move goods across the country.
Our “Waabi” Moment
Many of us can recount our first ride “Waymo moment” — sitting in the back of a driverless Jaguar I-Pace and realizing that autonomy has finally crossed over from research to reality.
Not too long ago, the G2 team had its own “Waabi moment” in Dallas, where we experienced a Class-8 truck flawlessly navigate highways, surface streets, and construction zones in moderate traffic. The system handled erratic drivers, merges, unprotected left turns, and even emergency vehicles smoothly and safely. The ride quality was calm and human-like, and, for lack of a better adjective, magical. What stood out the most was Waabi’s ability to achieve this performance at a tiny fraction of the cost and time versus its predecessors.
https://medium.com/media/c2b93a82ee34eebf65e7227c232bb8bf/href
Our Investment in Waabi
Since our initial investment in 2024, our conviction has only deepened that Waabi represents the next era of L4 autonomy: verifiable end-to-end AI capable of reasoning, simulation-first (grounded in realism), and a scalable architecture designed to work across long-haul trucking, robotaxis, and eventually personally-owned autonomy. What’s more is that the team has managed to consistently exceed our expectations.
Waabi is led by Founder & CEO Raquel Urtasun, a pioneer in AI, simulation, and autonomous driving, alongside COO Lior Ron, who most recently founded and scaled Uber Freight into a $5B business. This rare combination of deep technical leadership with commercial and operational expertise in freight is critical for commercializing autonomy.
That’s why we are thrilled to be co-leading Waabi’s Series C alongside Khosla Ventures and iconic strategic partners, including Uber and Nvidia (link). Our conviction stems primarily from Waabi’s differentiated technical approach coupled with impressive execution, as well as the structural opportunity in long-haul trucking.
Waabi’s Differentiated Approach
Waabi’s revolutionary Physical AI Platform enables true scale, generalizing to different form factors, geographies, and environments. This approach enables a shared brain across both autonomous trucks and robotaxis, in which the same AI model drives both applications. That means advances and progress results in improvements across both verticals.
Understanding how this is possible for the first time in the industry requires unpacking the technical strategy across three pillars: neural simulation, verifiable end-to-end AI capable of reasoning, and a rigorous safety architecture.
[1] Best-in-class neural simulator (Waabi World and Mixed Reality Testing):
Waabi World
Simulation has always been critical to autonomous development. Relying heavily on real-world miles is slow (rare edge cases take years to encounter, especially in trucking), costly (requires massive test fleets), and incomplete (real-world driving cannot expose systems to the full statistical distribution of scenarios).
While nearly every autonomy player we are aware of uses simulation to some degree, most take a “simulation-last” approach — testing primarily on public roads and using simulation for limited verification. Experts have described existing simulators as brittle, computationally expensive, and not end-to-end. They often simulate parts of the stack (e.g., motion planning), excluding perception entirely. As a result, most players still need to drive many millions of real-world miles to validate performance.
Waabi flips this paradigm.
Waabi World is a simulation-first system where offline simulation drives development and real-world driving is used primarily for fine-tuning. Its core attributes include:
- Physically Inspired Generative AI, reconstructing real-world geometry and appearance, from sensor data
- End-to-end closed-loop simulation, covering perception, prediction, planning, and control
- 99.7% realism score, making synthetic environments nearly indistinguishable from real roads
- Capital efficiency, replacing massive physical fleets with AI-driven validation
Crucially, Waabi does not treat simulation realism as a given. They have found a way to measure and prove it.
Waabi validates sim realism through a technique called “pair setting,” recreating real-world driving scenarios inside the simulator with matched actors, behaviors, lighting, weather, and road conditions. The autonomy system is then run in both environments using the same software release, and the resulting vehicle trajectories are compared directly. A 99.7% realism score indicates the system behaves nearly identically in simulation and the real world, unlocking simulation as a credible tool for safety validation, not just development. Read more about Waabi World and its realism benchmark here.
Mixed Reality Testing [MRT]
https://medium.com/media/df77a2b6d519333da8c4296c6c1bd075/href
MRT is a revolutionary technology that completely transforms what’s possible with closed-course testing, offering for the first time in the industry a way to test safety critical situations realistically and at scale.
In the same way that augmented reality goggles blend the physical world with a virtual world, MRT enables the Waabi Driver to drive autonomously down a physical test track while simultaneously experiencing numerous intelligent, simulated actors that coexist in this hybrid reality and react to each other and to the physical world in naturalistic ways. All this is possible by leveraging Onboard Waabi World, a version of Waabi’s neural simulator that runs in a few milliseconds on the onboard compute. As Onboard Waabi World generates new scenarios, the real physical sensor readings are modified instantaneously so the Waabi Driver can react to the blend of real and virtual elements while driving in the physical world. This fusion creates a first-of-its-kind reality that unlocks unlimited testing possibilities previously impossible to achieve safely or practically.
[2] Verifiable End-to-End AI (“next generation AV 2.0”)
For the past decade, autonomy has relied on an “AV 1.0” modular architecture — a brittle combination of rules-based systems and supervised learning split across perception, mapping, prediction, and planning. This approach is:
- Brittle → small changes cascade across modules
- Slow → feature updates take months or years
- Capital-intensive → requires billions of dollars and massive teams
- Hard to scale → errors compound across the stack
Waabi replaces modular stacks with a world model, analogous to how LLMs replaced hand-engineered language rules.
This architecture reasons about the world holistically, rather than stitching together modules. It generalizes more effectively, reduces compounding errors, and produces representations that are interpretable and verifiable for safety-critical deployment.
This model is what powers The Waabi Driver, providing true generalization that can safely and quickly scale autonomous driving across very different ODDs (operational design domains).
[3] Robust, disciplined safety architecture
In autonomy, performance metrics like “disengagements” or ride “smoothness” are often mistaken for safety. True safety is harder to prove and requires a different set of competencies than those needed to build highly performant AI.
What really matters is how the system behaves when things go wrong: slowing down and defaulting to caution in ambiguous situations, or executing minimal-risk maneuvers such as a safe pull-over. Waabi’s safety organization, led by André Strobel (former Waymo trucking safety lead), prioritizes functional safety, hazard analysis, standards-based validation, statistical behavioral analysis, and safety governance. This rigorous methodology is empowered by Waabi World as well as Mixed Reality Testing, enabling Waabi to better and more efficiently assess safety than anyone else in the industry
This discipline is critical to earning trust with regulators, OEM partners like Volvo Trucks, and customers. Our view is that trust and safety will always be a long-term differentiator in this industry.
Proven Execution in Autonomous Trucking
With a system that can be validated in simulation, generalized across environments, and deployed safely at scale, the question shifts from can autonomy work to where it creates the most immediate value. For Waabi, that answer is long-haul trucking.
Taken together, Waabi’s strategy has been backed up by impressive execution. Waabi has made significant strides to deliver next-generation autonomous technology and a commercial model that meets customer needs.
Some key milestones accomplished since the last round include:
- Running commercial operations with best-in-class fleets, retailers and shippers
- Successful completion of autonomy missions on a closed-course with no human onboard
- Validation of Waabi World with a 99.7% sim realism score
- Deploying mixed-reality for edge-case testing, pushing significantly the state-of-the-art in safety assessment.
- Successful integration of the Waabi Driver on Volvo’s fully-redundant autonomous truck.
- The ability to drive autonomously on both highways and generalized surface streets, which enables a Direct-to-customer business model where autonomous hubs are not required, and freight can be moved between end customer facilities, providing for the first time a product that meets the customer needs.
Why Trucking Is Ripe for Automation
Since inception, Waabi has largely been focused on commercializing long-haul trucking — a $1T industry in the U.S.¹ (measured in gross freight revenue) that is structurally broken across three vectors:
- Driver shortage: The U.S. faces an 80K truck driver shortage today, projected to double to 160K by 2030. Annual turnover is ~89%, creating a chronic threat to logistics continuity.²
- Safety risks: Trucking is one of the most dangerous professions, with 500K+ accidents and ~5K fatalities annually — 90% caused by human error.³
- Escalating costs: Over the past decade, the average marginal cost per mile has surged 43% ($1.58 in 2015 → $2.26 in 2024), with labor ($0.63 in 2015 → $1.00 in 2024) as the dominant driver of inflation.⁴
Autonomy offers a deflationary solution to the driver shortage, paired with meaningful operational savings from enhanced safety and efficiency.
Long-Haul Trucking Cost Per Mile (Illustrative)
Beyond cost savings, autonomy can fundamentally change asset utilization. Today, long-haul trucks are driven ~80K miles per year⁵, constrained by hours-of-service regulations, driver availability, and turnover. Autonomous trucks are not bound by these limitations, enabling materially higher utilization — potentially as high as 250K miles per year over time.
The combined effect of lower cost per mile and dramatically higher utilization transforms carrier economics, enabling up to a 10x increase in annual profit per truck.
Carrier Economics (Illustrative Present Day)
What attracted us to Waabi’s model is that its autonomy stack is designed for true door-to-door operations, with full capability on both highways and surface streets. Historically, most autonomous trucking efforts have focused on a hub-to-hub model — moving freight only between dedicated AV terminals and relying on human drivers for first- and last-mile delivery. This approach introduces operational friction, increases cost, and constrains customer adoption. By contrast, Waabi’s door-to-door capability eliminates handoffs, simplifies operations for carriers and shippers, and materially expands the addressable market.
Climate Impact
Autonomy has the potential to materially decarbonize trucking as it scales to widespread adoption. Autonomous trucks have the potential to deliver ~10–20% fuel-efficiency gains through smoother driving, optimized speeds, and reduced idling. At scale, autonomy can further reduce emissions by materially cutting “deadhead miles” (miles a truck drives without carrying cargo due to repositioning). Today, deadhead miles account for 35% of total trucking mileage. By enabling better routing, continuous operation, and tighter network coordination, autonomy could reduce deadhead miles by ~60% — representing 32B miles annually in the U.S.⁶
Autonomy is also likely to accelerate the transition to all-electric trucks: superior total cost of ownership, driven by lower energy cost per mile, higher utilization, and faster paybacks, unlocks widespread EV adoption, adding ~420 megatons in annual CO2e reduction by 2038.
Taken together, full penetration of AVs in trucking could abate ~0.57 GT CO₂e each year in the U.S. alone — roughly 85% of trucking’s emissions — with multi-gigaton potential globally. For context, total U.S. transportation emissions are ~1.9 GT annually⁷, meaning autonomous trucking alone could reduce emissions by 30% of the entire U.S. transportation sector. This is equivalent to removing ~128 million gasoline-powered cars from the road⁸.
Sources:
¹ Bloom Trucks,The Future of Freight, https://www.bloomtrucks.com/the-future-of-freight
² ATA, Driver Shortage Report and Forecast, https://www.trucking.org/news-insights/ata-releases-updated-driver-shortage-report-and-forecast
³ FMCSA, Large Truck and Bus Crash Facts 2022, https://www.fmcsa.dot.gov/safety/data-and-statistics/large-truck-and-bus-crash-facts-2022-1
⁴ ATRI, Operatoinal Costs of Trucking 2025, https://truckingresearch.org/wp-content/uploads/2025/07/ATRI-Operational-Costs-of-Trucking-07-2025.pdf
⁵ National Private Truck Council, 2025 Benchmarking Report, https://www.nptc.org/benchmarking/benchmarking-report/
⁶ ATRI, Operatoinal Costs of Trucking 2025, https://truckingresearch.org/wp-content/uploads/2025/07/ATRI-Operational-Costs-of-Trucking-07-2025.pdf
⁷ EPA, Transportation Sector Emissions, https://www.epa.gov/ghgemissions/transportation-sector-emissions#:~:text=In%202022%2C%20direct%20and%20indirect,to%20increased%20demand%20for%20travel.
⁸ EPA, Greenhouse Gas Emissions From a Passenger Vehicle, https://www.epa.gov/greenvehicles/greenhouse-gas-emissions-typical-passenger-vehicle
Why We Invested: Waabi was originally published in G2 Insights on Medium, where people are continuing the conversation by highlighting and responding to this story.