Autonomous driving and robotics technologies are part of a $200 billion market reshaping the automotive industry. Many tech companies are competing to dominate this market by trying to solve the autonomy challenges that self-driving vehicles have in urban environments.
A Mathematician’s Passion for Rock Climbing May Solve Autonomy Challenges
A trained mathematician, Vladislav “Vlad” Voroninski, serves as the co-founder and CEO of Helm.ai, one of the world’s leading providers of advanced AI-driven autonomous driving and robotics automation software. His objective is to create the most scalable and dependable AI software for autonomous driving and robotics.
Helm.ai currently works with some of the largest international OEMs, such as Volkswagen and Honda, to deliver innovative driver-assistance and autonomy systems to the consumer automotive market. “The industry has reached a tipping point where brute-force data collection is no longer commercially viable for high-end autonomy,” said Voroinski.
While others solely rely on data to improve autonomy, Voroninski uses a different approach inspired by his love for rock climbing. He wants to understand its structure by conducting an assessment of the “terrain,” isolating the required moves, and adapting them in real time. The idea is to change the way in which AI interprets the physical environment surrounding it.
Overcoming the Data Wall
The early development of autonomous self-driving systems depended on accumulating thousands of hours of real-world driving data to train the AI models. The theory was that the more data used to train the AI models, the more likely the AI would be able to respond well to critical “edge-case” scenarios, which are extreme circumstances that fall outside of normal parameters.
Unfortunately, the automotive industry has hit a “Data Wall,” where traditional autonomous driving AI models need rarer, more expensive data to improve their performance in edge-case scenarios. Some developers are turning to monolithic “end-to-end” models, but the problem is that they function as “black boxes.” They don’t have the interpretability required to qualify for the strict safety certification required for a Level 3 classification.
The Society of Automotive Engineers (SAE) International created a standard classification system to describe precisely how much a car can drive itself versus how much a human driver needs to operate it. Here is a brief rundown of the levels:
- Level 0 – No automation; All human driving
- Level 1 – Driver assistance only
- Level 2 – Partial automation; Driver must still monitor the road
- Level 3 – Conditional automation; Driver can take “eyes off” the road under specific driving conditions only
- Level 4 – High Automation: The car can operate on its own in specific, mapped-out urban areas.
- Level 5 – Full Automation: The car can drive itself anywhere in all conditions and locations without human assistance.
The ultimate goal of companies like Helm.ai is to achieve Level 5. Helm.ai recently announced a significant capability expansion for its production-ready, vision-only software stack, Helm.ai Driver, designed to scale effortlessly from Level 2 through Levels 3 and 4. The Helm.ai Driver system is built on the company’s Factored Embodied AI architecture to enable human-like autonomous driving in urban traffic without needing a lidar sensor and high-definition maps.
How the Factored Embodied AI Works
The Factored Embodied AI allows autonomous systems to use far less training data to make critical decisions. In 2025, Helm.ai launched an urban pilot program to test its AI Driver system in a busy city area. As it turned out, the AI Driver was able to achieve vision-only, zero-shot autonomous steering while navigating the complex city streets of Torrance, California.
The AI Driver system interprets the common urban autonomy problem as a matter of “Perception” and “Policy” to distinguish between road geometry and traffic rules. Automotive OEMs require such transparency and reasoning in their autonomous software to scale from a Level 2 deployment to Level 3 and 4 deployments.
“By delivering a vision-first system that powers advanced Level 2+ today, and serves as the software brain for the transition to Level 3 and Level 4 autonomy, we are providing OEMs with the only realistic path to deploying next-generation autonomy on mass-market compute platforms,” said Voroninski.
Achieving urban capability for autonomous systems would normally require OEMs to spend billions of dollars and require millions of miles of training data. However, Helm.ai Driver was able to achieve training maturity from using only 1,000 hours of real-world driving data, thanks to a proprietary unsupervised learning technique called Deep Teaching.
Deep Teaching enables neural networks to learn directly from massive amounts of data. By pairing this technique with advanced semantic simulation, the AI Driver system can quickly train on infinite geometric scenarios rather than needing to render endless amounts of photorealistic raw pixels. That significantly lowers the time and costs associated with autonomous development.
Can It Handle All Geographies?
The real test will be when OEMs produce massive numbers of vehicles with this new autonomous system technology in them and distribute them to different geographies throughout the country. Since the autonomous system will be exposed to new environments without HD maps and manual tuning, it will need to successfully navigate these unseen areas.
When Helm.ai deployed its AI Driver in the city of Torrance in the Greater Los Angeles Area, it did not have any prior training of the area’s specific streets. It was able to execute “zero-shot” autonomous driving by generalizing the surrounding geographies. The successful pilot test showed that Helm.ai’s international OEM partners can scale up to Level 4 without bearing the high cost of geofencing and collecting data for each city.
About Helm.ai
Helm.ai was founded in 2016 for the purpose of developing AI software for Advanced Driver Assistance Systems (ADAS), robotics automation, and autonomous driving. Under Voroninski’s leadership as CEO, Helm.ai has managed to raise approximately $103 million in funding to expand its autonomous driving and robotics programs and strengthen its partnerships with international automakers and OEMs.
Voroninski is a graduate of UC Berkeley with a Ph.D. in Mathematics and a graduate of UCLA with a Bachelor of Science and Master of Arts in Applied Mathematics. His work background includes serving on MIT’s mathematics faculty and being the Chief Scientist who founded Sift Security, a machine-learning cybersecurity company.
