Almost two years ago, Motional found itself at a pivotal moment in the development of autonomous vehicles.
The company, backed by a $4 billion investment from Hyundai Motor Group and Aptiv, had missed its deadline with partner Lyft to launch a driverless taxi service. With Aptiv stepping back as a financial backer, Hyundai increased its commitment with a fresh $1 billion investment to keep the company afloat. Multiple rounds of layoffs followed, including a major 40% workforce reduction in May 2024 that cut headcount from around 1,400 to under 600. At the same time, rapid advancements in AI were reshaping how engineers approached technology development.
Motional faced a stark choice: evolve or risk obsolescence. It chose transformation—pausing operations to rebuild with a new strategy.
Motional has now restarted its robotaxi program with an AI-first approach to developing its self-driving system, aiming to launch commercial driverless services in Las Vegas by the end of 2026. The company is already offering employees access to a ride-hailing service using autonomous vehicles—each equipped with a safety operator on board. Later this year, it plans to open the service to the public in partnership with an unnamed ridesharing platform (Motional maintains existing collaborations with both Lyft and Uber). By year-end, safety operators are expected to be removed entirely, marking the start of fully commercial, unsupervised autonomous operations, according to the company.
"We’ve seen incredible progress in AI that brings enormous potential," said Laura Major, President and CEO of Motional, speaking at the company’s Las Vegas facility. "At the same time, while we had a safe autonomous system, there was a gap in achieving an economically viable solution that could scale globally and broadly. That led us to make the difficult decision to pause commercial activities—slow down temporarily so we can accelerate later."
This shift involves moving away from traditional robotics methods toward an AI foundation model architecture. Motional never lacked AI capabilities; its earlier system used discrete machine learning models for perception, tracking, and semantic reasoning. However, other parts of the software stack relied heavily on rule-based programming. As Major explained, this patchwork of separate models created a complex and rigid software ecosystem.
In parallel, AI architectures originally developed for natural language processing—particularly transformer models—began proving effective in robotics and physical AI systems, including autonomous driving. These breakthroughs enabled large-scale AI models like ChatGPT and opened new pathways for end-to-end learning in real-world environments.
Motional is now integrating those smaller, specialized models into a unified backbone, enabling a more cohesive, end-to-end system. Smaller models remain available for specific developer tasks, giving the company the best of both worlds—scalability and flexibility.
"This is critical for two reasons," Major said. "First, it enables much better generalization across new cities, environments, and scenarios. Second, it allows us to do so in a cost-efficient way. For example, traffic signals in the next city might look different, but you don’t need to re-engineer or re-analyze everything. You simply collect some data, fine-tune the model, and it can operate safely in that new location."
TechCrunch took a 30-minute autonomous ride in Las Vegas to experience Motional’s updated approach firsthand. While a single demo cannot fully evaluate a self-driving system, it offers insights into key improvements over previous versions.
One notable advancement occurred when the Hyundai Ioniq 5 I was riding in autonomously exited the Las Vegas Strip and navigated into the pickup zone at Aria Resort & Casino. These areas are notoriously chaotic, and my ride was no exception—handling slow maneuvers around stopped taxis, passenger drop-offs, lane changes, and dense foot traffic amid large planters and vehicles.
Prior to this update, Motional operated a ride-hailing service in Las Vegas with partner Lyft, where vehicles handled certain segments autonomously. However, parking lots, valet zones, and crowded hotel curbsides were always managed by human safety drivers. These complex micro-environments were beyond the scope of automated control.
There is still room for improvement. The in-vehicle passenger interface remains under development. While my demonstration ride experienced no disengagements—meaning the safety operator never needed to take control—the vehicle did pause briefly when navigating around a double-parked Amazon delivery van.
Still, Major believes Motional is advancing safely and efficiently. She emphasized that its majority owner, Hyundai Motor Group, remains committed for the long term.
"The ultimate long-term vision," Major said, "is applying Level 4 autonomy—where the system handles all driving without human intervention—to personal vehicles. Robotaxis are just the first step, delivering significant impact today. But eventually, every automaker will want this technology embedded in their cars."
