Swim Swarm: UAVs & Machine Learning at the Edge

by Brad Johnson, on Sep 19, 2017 11:33:59 AM

Swim Swarm demonstrates how Swim ESPTM for Autonomous Vehicles can be used to transform devices such as Unmanned Aerial Vehicles (UAVs) from low-value data into an affordable source of low-rate but high value insights, which are able to autonomously make fast-timescale decisions. Swim also controls the UAVs as they execute their mission as a swarm.

Mike Gualtieri, of Forrester (May 2016), states that streaming analytics applications "...need a new, stream-driven machine learning approach. Sometimes data spikes fast and furiously, and at other, it drips in.

  • IOT apps must be able to ingest data at the rate it is emitted.
  • Streaming apps must be able to analyze data in real-time.
  • Platforms need small footprint versions that can perform ingestion and analytics at the edge."

Streaming analytics apps filter, aggregate, enrich and analyze the content of high throughput flows of ephemeral data to permit an automated of human response on a time-scale similar to that of the data itself.

Analyst firm Forrester Research Inc. has identified key challenges for organizations seeking to deliver learned insights from their sensor-rich environments:

  • Whereas legacy applications use a “request and response” approach in which apps get input, process requests, and then respond, this doesn’t work for autonomous vehicle applications because they must be ready to adapt at any time.
  • Moreover, many new applications need to learn from data immediately, to deliver insights in context and in real-time to allow operators or applications to respond rapidly to a changing environment.
  • Finally, it is unrealistic to expect high quality labeled data with which to train the Machine Learning application. UAV operators lack the expertise for this, so Machine Learning algorithms need to be robust to noisy data and they need to be self-training.

UAV use cases impose complex new requirements for processing data flows at the edge:

  • Transformation: Streaming data delivers tremendous volumes of noisy information. Transformation narrows the stream, reducing bandwidth by including only data relevant to an application.
  • Correlation: UAV applications combine data from multiple sensors. Correlation permits applications to identify events that are related in time, space or via some other causal link, helping operators to identify and address underlying causes. The system must be stateful and able to correlate observations from different sensors.
  • Enrichment: Sensor data has no contextual reference to the mission. Through integration with mission control databases, the context-linked enriched data becomes mission relevant.
  • Time: Streaming data flows in real time, but applications need to transform this into observations relevant to various time windows for time-series analysis.
  • Mission logic: The goal of stream analysis is to inform applications with real-time context at a high level of semantic value, so that decisions can be made – by operators or automatically. This eliminates the “programming” by enterprise developers.

Learn how SWIM implements Machine Learning at the edge to transform sensor data generated by UAVs from low-value data into low-overhead, high value insights capable of informing fast-timescale decisions by UAVs.

Topics:Machine LearningSWIM SoftwareIndustrial IOTEdge AnalyticsdronesUAV'sSWIM Swarm