If you use Jupyter notebooks to explore and analyze your robotics data (or want to learn how to!), check out our step-by-step tutorial below:
We stream nuScenes self-driving car data (stored in Foxglove Data Platform) into our Jupyter Notebook, and then write code with data analysis libraries like pandas and matplotlib to render a few basic visualizations.
Through these visualizations, we’re able to surface a few meaningful insights about our robot’s data – the route it took, its acceleration throughout its drive, and the perceived objects it encountered on the road.
We’re planning to publish a follow-up post expanding on this dataset – please let us know if you want to learn how to accomplish any particular tasks!
That’s looking nice, exactly the same as how one would acquire data from a bag.
When I read this, I immediately wondered how to handle very large amounts of data, since pandas is in-memory, but your data-store could potentially contain very much data. The selection statements go some way towards this, by selecting just a subset of what’s available, but still.
Are you planning to support dask or similar things?
Hi, thanks for the question. We are planning to add support for distributed compute, though most of our initial thinking has been around Apache Spark, for Databricks interop and availability of managed Spark mainly. We are not familiar with dask, although I do see positive reviews.
Any insights or feature requests you may have in this area are very welcome!
