Mighty AI tackles training data for autonomous driving
Seattle-based self-driving car startup Mighty AI, which helps automotive companies get the labeled datasets they need to train their autonomous vehicles to see, is opening an office in Detroit. The company, previously known as Spare5, is a venture-funded startup that uses data to train computer vision algorithms to name and detect objects, describe attributes, and associate relationships and interactions. It is backed with funding by six institutional investors: The Foundry Group, Madrona Venture Labs, NEA (New Enterprise Associates), Accenture, Intel Capital, and GV (formerly Google Ventures).
“Mighty AI is all in on autonomous driving,” said Matt Bencke, Founder and CEO, Mighty AI, in a company blog post. “And I’m thrilled to share that, as we double down on delivering training data for autonomous vehicles, we’re going where the action is… We’re opening a new location in Detroit.”
Mighty AI is working with several of the world’s leading OEMs, suppliers, and startups, to deliver training data for advanced driver-assistance systems. It will be working out of a new space in downtown Detroit provided by Techstars Mobility, which invests globally in startups transforming how people and goods move around, spanning mobility categories including automotive, trucking, transportation, smart city, energy, logistics, supply chain, and travel.
“We hear from automakers all the time that not having accurately labeled data is the number one blocker to getting to higher levels of autonomy. This is especially true when you start talking in terms of the petabytes of data that needs to be labeled,” said Angela Cherry, Director of PR and Communications, Mighty AI.
According to Bencke, the nearly three-year-old company employs 45 people that have custom built a platform and team to provide training data as a service, “Using [our] community of over 200,000 people around the world who use Spare5, and we push them custom-designed microtasks, we use our own machine learning to mathematically predict every single microtask (who in our community is going to complete it accurately), we work very hard on the tooling to make sure it is easy and fun for them to do that, then we use our machine learning to process for quality. We integrate with our customers and help them train their models for things like sentiment analysis or speech recognition, and variety of use cases around computer vision.”
Specifically, Mighty AI’s suite of annotation tools used to train AV models include detection and localization; polygons for road boundaries, road markings, and Botts’ dots; cuboids for vehicles and side panels; categorizations for destination path, next road, weather, traffic, pedestrians, speed, topology, and danger; classifications on object saliency and visibility; directional commands for driving scenes; associations for body and head orientation metadata; key point annotations for pedestrian joints; and segmentation masks.
The aim is to accelerate the development of highly automated vehicles. For more info on how Mighty AI works with training data for autonomous driving, visit here.