BREAKING ● Roboto AI turns petabytes of robot logs into a search box Seattle HQ ● spun out of the Allen Institute for AI $4.8M seed led by Unusual Ventures, AI2 & FUSE ROS · PX4 · MCAP · Parquet ● one Python SDK Founded by ex-Amazon Robotics engineers Named a Top 50 Most Innovative Robotics Company BREAKING ● Roboto AI turns petabytes of robot logs into a search box Seattle HQ ● spun out of the Allen Institute for AI $4.8M seed led by Unusual Ventures, AI2 & FUSE ROS · PX4 · MCAP · Parquet ● one Python SDK Founded by ex-Amazon Robotics engineers Named a Top 50 Most Innovative Robotics Company

Company Dossier — Robotics Data Infrastructure

Roboto AI reads the logs nobody has time to read.

The analytics engine for Physical AI - built by ex-Amazon drone engineers to find the one edge case buried in a fleet's worth of sensor data.

Seattle, WA · est. 2022

THE MARK: A wordmark where the dotted "i" reads like a robot's single, watchful eye - and four little shapes that stand in for four messy data formats about to be made searchable.

The Scene

It's 2am, and somewhere a robot did something weird.

A drone clipped a gust it shouldn't have. A warehouse arm hesitated for 40 milliseconds. A quadruped's foot slipped on a floor it has crossed a thousand times. None of it made the news. All of it is sitting, right now, in a log file - one of millions - inside a bucket the size of a small city.

Somewhere, an engineer is scrolling. Not debugging, exactly. Scrolling. Squinting at a timeline of raw telemetry, hoping the anomaly announces itself. This is how most robots get fixed: by hand, by hunch, by the person who happened to remember which folder to open.

Roboto AI exists because that scene is absurd. Robots produce petabytes. Humans read one screen at a time. The gap between those two facts is where good robots quietly become unreliable ones.

So a small team in Seattle asked an unglamorous question - not "how do we build a smarter robot," but "why can't anyone search the robots they already have?" - and built the answer into a platform.

What It Actually Does

Raw recordings in. Answers out.

Every robotics team speaks a different data dialect. Roboto speaks all of them. It ingests the mess, indexes everything, runs automated health checks and AI analysis agents, then hands back the moment that matters instead of a timeline to scroll.

ROS bags

PX4 / ULog

MCAP

Parquet

Video

→

ROBOTO ENGINE · index + AI agents

→

Edge cases · anomalies · datasets

01 / SEARCH

Fleet-Wide Search

Query patterns across entire fleets and millions of miles of data - tags, metadata and topic statistics combined into one search.

02 / DETECT

Event Detection

Automatically flag vibration spikes, GPS jamming, sensor failures and anomalies before they reach the field.

03 / BUILD

Python SDK & CLI

Pull robotics logs straight into pandas DataFrames and automate workflows over terabyte-scale sensor data.

04 / CURATE

Dataset Curation

Turn raw recordings into structured datasets for training and evaluation - foundation-model-ready.

05 / MEASURE

Custom Metrics

Track reliability and performance KPIs across a fleet, with time-synced playback of video, telemetry and logs.

06 / SECURE

Enterprise-Ready

Role-based access control, data retention policies, multi-user collaboration and private cloud deployment.

The Founders

Two people who met building robots - and got tired of the aftermath.

Benji Barash

Co-founder & CEO

Former engineering leader at Amazon Robotics and a founding engineer on Amazon's UK drone delivery program in Cambridge, where he built large-scale simulation platforms and safety-critical autonomy algorithms. He kept hitting the same wall: nobody could search their own logs.

Yves Albers-Schoenberg

Co-founder & CTO

Former science leader at Amazon Robotics and AWS, and a computer-vision researcher out of ETH Zurich with work published at IROS and ICRA. The scientist to Benji's builder - and the reason "search by signal similarity" isn't marketing copy.

Who's On Board

From drones to surgical robots.

The teams building the physical future are drowning in the exhaust of building it. A few of the ones reaching for a snorkel:

BRINC

ANYbotics

Telos Health

Cleo Robotics

Drone makers · medical & surgical robotics · autonomous vehicles · general robotics teams, prototype to production.

The Paper Trail

How Roboto got here.

2022

Roboto spins out of the Allen Institute for AI (AI2) in Seattle.

APR 2023

Emerges from stealth with a $4.8M seed round led by Unusual Ventures, with AI2 and FUSE Ventures.

APR 2025

Benji Barash appears on theCUBE for NYSE Wired Robotics & AI Media Week.

JUN 2025

Live demo of AI agents debugging robotics data at the Cloud Robotics Working Group.

Marginalia

Things worth dog-earing.

The eye

The dotted "i" in the Roboto wordmark reads like a robot's single, watchful eye.

The origin

The founders met building drones and warehouse robots at Amazon - then decided the real problem was the data those robots left behind.

The inbox test

The platform reads robot logs the way your inbox reads email: petabytes collapse into a search box.

Five dialects

ROS, PX4/ULog, MCAP, Parquet and raw video - all ingested through one pipeline.

Back To The Scene

It's 2am. The robot did something weird. This time, the data speaks first.

No one is scrolling. The anomaly - a vibration spike three flights ago, a signal that looked almost like a dozen others across the fleet - has already surfaced itself, already been matched to its near-twins, already been folded into a dataset the team can train against by morning.

The engineer isn't hunting anymore. She's asking questions and getting answers, in plain language, from a fleet she can finally hold in one hand. The robot is still weird. But now weird is legible - and legible is fixable. That is the small, stubborn, unglamorous thing Roboto AI changed: it made the exhaust of the robotics revolution something you can actually read.