Two years ago, Motional faced a critical juncture in its autonomous vehicle journey. The company, established through a $4 billion joint venture between Hyundai Motor Group and Aptiv, had failed to meet its initial deadline for launching a driverless robotaxi service with Lyft. After Aptiv withdrew its financial backing, Hyundai provided an additional $1 billion investment to sustain operations. A series of workforce reductions, including a major 40% cut in May 2024, reduced employee numbers from a peak of roughly 1,400 to under 600. Concurrently, rapid progress in artificial intelligence was reshaping how self-driving technology was engineered. Confronted with the imperative to adapt or cease operations, Motional chose to evolve, putting its commercial plans on hold.

The company has since initiated a robotaxi service for its employees, which currently includes a human safety driver. Plans are in place to launch this service publicly through an undisclosed ride-hailing partner later this year. Motional has existing partnerships with both Lyft and Uber. According to company statements, the safety operator will be removed from the vehicles by year’s end, marking the start of a fully commercial, driverless service.

“We recognized the tremendous potential created by recent AI advancements. While we already had a safe, driverless system, a gap remained in achieving an affordable solution capable of global generalization and scaling,” explained Motional president and CEO Laura Major during a presentation at the company’s Las Vegas facilities. “Consequently, we made the difficult decision to pause commercial activities, slowing down in the short term to accelerate our progress in the long run.”

This strategic shift involved moving from a traditional robotics methodology to an approach centered on AI foundation models. Motional’s self-driving system had always incorporated AI, utilizing separate machine learning models for perception, tracking, and semantic reasoning, complemented by rules-based programming for other functions. According to Major, this structure resulted in a complex and intricate software framework.

The emergence of transformer architectures, originally developed for language processing and later applied to robotics and autonomous driving, enabled the creation of large, sophisticated AI models—exemplified by the rise of ChatGPT. Motional sought to consolidate its numerous smaller models into a unified, end-to-end architectural backbone. The company has retained these individual models for developer use, which Major describes as offering the best of both approaches.

“This integration is crucial for two reasons: it allows for easier generalization to new cities, environments, and scenarios, and it enables cost-optimized scaling,” she noted. “For instance, traffic light systems may differ in a new city, but instead of redeveloping the software, you can simply collect data, train the model, and achieve safe operation there.”

While a single demonstration cannot fully evaluate a self-driving system, it can reveal specific weaknesses, highlight improvements from prior versions, and indicate overall progress. Such progress was evident during a ride in a Hyundai Ioniq 5, which navigated autonomously off Las Vegas Boulevard into the busy pickup and drop-off zone at the Aria Hotel. In this notoriously chaotic area, the vehicle cautiously maneuvered around a stationary taxi and disembarking passengers, executed lane changes, and passed numerous pedestrians, large planters, and other cars.

Previously, Motional and Lyft operated a ride-hailing service in Las Vegas where vehicles autonomously handled only certain segments of a trip. Complex areas like parking lots and hotel valet zones were excluded; a human safety driver would always take control in those situations. Further advancements are still necessary. The in-vehicle rider display graphics remain in development, and during the demo, the autonomous vehicle, while never requiring a safety driver intervention, hesitated noticeably when navigating around a double-parked Amazon delivery van.

Nevertheless, Major asserts that Motional is on the correct trajectory to achieve safe and cost-effective deployment, with majority owner Hyundai committed for the long term. “The ultimate long-term vision is to implement Level 4 automation into personal vehicles,” Major stated, referring to systems that operate entirely without human intervention. “Robotaxis are the first major step and will have a significant impact. But ultimately, any automotive manufacturer would aspire to integrate this technology into their consumer cars as well.”