Our first release enables robotics engineers to manage and transform sensor data. Think Dropbox meets GitHub Actions!
October 18, 2023
The field of robotics is rapidly expanding and so is the proliferation of massive amounts of multi-modal sensor data. Yet, existing data platforms like Splunk or Datadog fall short, lacking support for specialized robotics and sensor data formats like ROS bags and PX4 log files. This gap forces engineers to spend way too much time crafting custom data ingestion and aggregation solutions instead of focusing on their core products. These makeshift solutions are not only cumbersome and costly to build and maintain, but also fail to scale, creating a massive divide between industry leaders like Tesla and emerging robotics teams.
It’s all about the foundation!
During our time at Amazon Robotics, we learned that a critical step for any robotics project is to establish a centralized data platform. Today, many teams fail to prioritize investments in data infrastructure due to resource constraints and complexity. This leads to a data deluge as they expand and their fleets of robots grow.
Having seen many robotics projects fail to reach production, we advise teams to focus on data infrastructure as early in their development as possible and to ensure they prioritize the following core features:
Storage: Picture Dropbox, but with first-class support for sensor data. A central data repository ensures universal access across teams, streamlining the process to retrieve and analyze relevant data quickly.
Transformation: Picture GitHub Actions, but tailored for sensor data. A transformation engine must be robust and flexible, and able to power distinct workflows across different teams, such as Perception and Path Planning.
Collaboration: Robotics is an interdisciplinary field. Platforms must encourage data sharing, tagging and commenting. Seamless collaboration between teams accelerates development and increases system reliability and safety.
Implementing this foundational data infrastructure early creates a flywheel effect, significantly increasing development velocity. With a core data storage and transformation engine, teams can leverage advanced analytics—like signal search and anomaly detection—on top of it.
Efficient ingestion and processing of large-scale, multi-modal sensor datasets is also crucial to capitalize on the latest advances in Large Multimodal Models (LMMs) and other types of foundation models that are being released.
We're currently at ROSCon 2023 in New Orleans, and very excited to release a developer preview of Roboto's foundational data platform which provides all of the key features above. If you're building robots, or work with lots of sensor data, sign up for free and try it out. We’d love to get your feedback!