Interested in learning what's next for the gaming industry? Join gaming executives to discuss emerging parts of the industry this October at GamesBeat Summit Next. Learn more., a startup creating software for autonomous vehicles, today announced that it raised $26 million in series B financing led by Amplo, JMPartners, Base Capital Funding, and Freeman Group with participation from ACVC, One Way Ventures, Binnacle Partners, and individual investors. According to CEO Vlad Voroninski, the funding will help Helm “better serve its customer base” while supporting product expansion.

Self-driving technologies have captured the attention of investors during the pandemic, particularly as strains on the supply chain — exacerbated by a driver shortage — make apparent the usefulness of automated delivery trucks. For example, earlier this year, self-driving software startup Embark announced plans to go public in a deal that would value the company at more than $5 billion. For its part, Alphabet’s Waymo, which is developing both driverless delivery trucks and taxis, closed a $2.5 billion funding round in June ahead of an expansion to San Francisco.

Helm, which is based in Menlo Park, California, was founded in November 2016 by Voroninski and Tudor Achim. A Russian-American mathematician and entrepreneur, Voroninski was the chief scientist at Sift Security, a cybersecurity machine learning startup, which was acquired by Netskope in 2018.


Transform 2022

Join us at the leading event on applied AI for enterprise business and technology decision makers in-person July 19 and virtually from July 20-28.

Register Here

“I started working in computer vision and AI 16 years ago during the DARPA Grand Challenges. I understood the bottleneck to AI to be an applied mathematics problem, and focused on math and its applications to machine learning and computer vision for about a decade in both academia and industry,” Voroninski told VentureBeat via email. “Several years ago, it became clear that AI will cause a massive transition in society via its transformation from the purely digital to the physical realm, including safety critical applications like self driving cars. It was the recognition of that impending shift and the necessity of huge advances in unsupervised learning to enable it which led to the founding of Helm.”

Enabling autonomous driving

Helm licenses AI software to perform perception, intent modeling, path planning, and control for autonomous vehicles. While this isn’t too far off from the approach competitors like Mobileye are taking, Helm isn’t selling its own sensor packages. Moreover, it’s leveraging a machine learning technique called unsupervised learning to ostensibly eliminate the need for large-scale driving data collection and annotation.

Autonomous vehicle systems ingest massive amounts of data to learn to drive without a human at the wheel. Embedded in the data are images of pedestrians, cars, signs, and other objects, most of which must be manually labeled so that the system learns to recognize them on their own.

Where labeled datasets don’t exist, unsupervised learning — also known as self-supervised learning — can help to fill the gaps in domain knowledge. For example, Facebook’s recently announced SEER, an unsupervised model, trained on a billion images to achieve state-of-the-art results on a range of computer vision benchmarks.

Helm’s system takes in data from dash cameras, processes the images, clusters them, and automatically tags them. Leveraging this technique, the company claims its full-stack solution can steer on steep and curvy mountain roads using only one camera without training on data from these roads.

In Voroninski’s words: “[Helm’s] technology leverages cutting-edge applied mathematics to unlock the full potential of deep learning for deploying safe autonomous systems at scale. We identified some key challenges that we felt weren’t being addressed with the traditional approaches, in particular regarding the scalability and accuracy of AI software.”

Unfortunately, unsupervised learning doesn’t eliminate the potential for bias or flaws in the system’s predictions. Some experts, including Facebook chief scientist Yann LeCun, theorize that removing these biases might require a specialized training of unsupervised models with additional, smaller datasets curated to “unteach” specific biases.

Thirty-employee Helm didn’t respond to a question about bias mitigation measures. However, the company did say that it’s working with “several of the top OEMs” in addition to Honda’s Xcelerator program and “chip companies and automotive Tier 1s.” In addition to this, Helm is also exploring applying its technology to other domains, including aviation, consumer and industrial robotics, construction and mining, retail automation, and health care.

“Despite the pandemic which has affected the automotive sector, Helm has seen a significant acceleration in business development and revenue in the last year … We have substantial revenue and revenue growth,” Voroninski told VentureBeat. “[Our mission remains achieving] autonomous driving and the automation of robotics at large, with an initial focus on building state of the art self-driving vehicle software using the power of unsupervised deep learning.”

VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Learn more about membership.