To create the autonomous technologies of the future, it’s essential to effectively manage extensive data collection, particularly video footage crucial for training AI models. Companies focusing on self-driving vehicles, robotics, and automated construction processes often amass countless hours of video footage for analysis. Traditionally, humans have been tasked with sifting through these vast archives for valuable insights, a method that is largely impractical.
NomadicML, co-founded by Mustafa Bal and Varun Krishnan, aims to revolutionise this process. The startup seeks to address the issue of underutilised fleet data, with over 95% often archived and forgotten. Their innovative platform leverages vision language models to transform raw video into structured, searchable datasets, enhancing the ability to monitor fleets more effectively and generating unique datasets for reinforcement learning.
Recently, Nomadic secured an impressive $8.4 million in seed funding, which bolsters its capacity to attract new customers and refine its offerings. The investment round, valuing the company at $50 million, was led by TQ Ventures with participation from Pear VC and noted AI expert Jeff Dean. Notably, the startup also clinched the top prize at Nvidia’s GTC pitch competition.
Bal and Krishnan’s journey began as Harvard undergraduates, where they encountered recurring technical challenges in their careers at firms like Lyft and Snowflake, leading to the inception of Nomadic. Their platform offers clients valuable insights from their own footage—a necessity for advancing the capabilities of autonomous vehicles and robotics. The technology can identify unique driving scenarios, such as navigating through unusual environments, facilitating both compliance tracking and integration into training pipelines.
Companies such as Zoox, Mitsubishi Electric, and Zendar have already begun utilising Nomadic’s tools to enhance their machine-learning operations. Zendar’s VP of Engineering, Antonio Puglielli, highlighted how the platform significantly accelerates working processes, distinguishing itself from conventional outsourcing approaches.
Nomadic’s model-based auto-annotation technology is positioning itself as pivotal in the field of physical AI, with competitors like Scale, Kognic, and Encord also developing associated AI tools. Additionally, Nvidia has launched the open-source Alpamayo models aimed at addressing similar challenges.
Varun envisions their platform as more than a mere labelling tool; it serves as an “agentic reasoning system” that autonomously identifies necessary data, understanding contextual nuances through multiple models. TQ Ventures’ partner, Schuster Tanger, likens Nomadic to foundational tech firms, asserting that any attempt by autonomous vehicle companies to replicate Nomadic’s functions internally might detract from their core mission of developing sophisticated robots.
The team at Nomadic, comprising talented engineers—including a world-ranked chess master—is dedicated to creating advanced tools, such as those that analyse lane change dynamics and improve robot precision in grasping objects. Looking ahead, Nomadic aims to develop solutions that address non-visual data, like lidar readings, and to integrate diverse sensor information. Bal aptly noted the complexity involved in managing terabytes of video alongside extensive models to derive actionable insights.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
