Log management is one of the most important tasks that every autonomy program needs to master. Test fleets collect on average four terabytes of drive data per vehicle per day, while production fleets (i.e., vehicles purchased by individual consumers) can generate millions of events per day. This firehose of data has enormous potential to power an autonomy program’s development efforts.
Due to the costs and risks involved in real-world testing, it is crucial that autonomy programs collect and manage their drive data effectively. For example, to operate a test fleet, autonomy programs must purchase and maintain vehicles and sensors and pay a team of safety operators. Additionally, just one critical mistake during real-world testing can put human lives at risk. Autonomy programs should thus implement practices to scale their data collection efficiently, create a pipeline for effective drive data processing, and build scalable workflows that extract the maximum value from all collected data.
Applied Intuition’s log management handbook discusses the technical building blocks, ideal workflows, and cost management strategies of an expansive drive data management process. This blog post is the first in a three-part series providing a short introduction to these topics. The full-length handbook is available for download below.
In autonomous systems development, log data is any real-world drive data collected on the system corresponding to the autonomous task at hand. For autonomous vehicles, log data is collected during a drive and ranges from raw sensor inputs to pedal or wheel actuation commands.
The concepts, principles, and approaches laid out in our log management handbook apply to autonomy programs of all sizes and across industries. Most metrics and examples concern SAE Level 2-4 systems in automotive, but the contents of our handbook are equally relevant to autonomous trucking, construction, mining, and agriculture as well as warehouse robots, unmanned aerial systems, and other types of autonomous systems.
Our handbook’s structure follows the journey of a drive data file from inception to long-term storage (Figure 1). First, an autonomous system collects the drive data file. Next, data processing pipelines distribute it, and different teams explore it according to their specific use case. Finally, the drive data file lands in long-term storage. Our handbook discusses each of these steps in detail.
Drive data powers various workflows for different teams within autonomous systems development. Our handbook covers the following workflows:
Drive data is one of the most essential building blocks of autonomous systems development. Our log management handbook explains common drive data management challenges and lays out recommended practices for autonomy programs across industries. The next part of this blog post series will summarize the handbook’s key insights regarding an important step in the log management life cycle: Drive data exploration. Stay tuned for the next blog post, or download the entire handbook today.