Building robotics platforms is a difficult business and requires processing tremendous amounts of data, but there are very few tools that help developers in that task. Roboto Technologies Inc., a startup that emerged from stealth today with $4.8 million in funding, aims to use artificial intelligence to help robotics developers save time when developing robots.

The seed round was led by early-stage investment firm Unusual Ventures, with participation from Allen Institute for Artificial Intelligence and FUSE Ventures.

Roboto founders Benji Barash and Yves Albers-Schoenberg saw firsthand the challenges of robotics development when they immigrated from Europe to Seattle and joined Amazon’s drone delivery program.

“We’ve had an amazing front-row seat to the incredible potential of robotics, having spent our careers working at Amazon on projects such as drone delivery,” the founders said in the announcement. “However, we also learned that, despite all the recent progress, building safe and reliable autonomous systems is still surprisingly hard and very expensive.”

A single robot can produce terabytes of complex data during mere minutes of operation, and this needs to be stored and processed to be useful. Many robotics platform developers must then create their own tools to process, understand and analyze it and the tools on the market are just not built to manage it.

For example, data and log parsers such as Splunk, Tableau and Datadog are great for dealing with unstructured data. But they’re not so great for video, images, lidar and other visual data coming from numerous sensors that robots use to understand the world around them. As a result, engineers spend hundreds or thousands of work hours building their own infrastructure to make existing tools fit robotics, rather than spending time working on robotics development itself.

“This is time-consuming and quickly becomes a pain point for everyone working with robots,” the founders said. “As companies collect giant volumes of data, robotics engineers have to spend their time wrangling it instead of doing robotics work.”

With its AI-powered platform, Roboto does away with these pain points by making it possible to analyze and fully understand tremendous amounts of sensor data without the need for homegrown tools.

Barash and Albers-Schoenberg likened the new system to a “copilot for robotics,” stating that developers could use it to discover anything they needed by asking it questions about their datasets. Using the AI, researchers could get information about their platforms as easily as performing searches using natural language prompts about sensor data, camera visuals or other inputs.

For example, a developer with an autonomous truck could ask the AI, “Find me all the drives where vehicle speed was greater than 35 mph and a person was seen on the right side of an image,” or “Show me every time a right turn happened and speed exceeded 15 mph.” All of these instances could be used to bring up the data associated with these instances in order to see what was happening for further investigation.

The platform also allows searches of other types of data such as graphical time-series signals, such as vehicle steering, battery level, speed and acceleration. By providing a way for robotics developers to search this sort of data rapidly, and using natural language prompts, developers can quickly understand how their robots are operating.

Roboto has released a free demo for developers to test the platform on data from nuScales, which is a large-scale autonomous dataset used by engineers and researchers.

Image: Roboto