[Nav2] Open Navigation's Robotics Workload Benchmark Release

Howdy everyone - Your Friendly Neighborhood Navigator here with one more key announcement this month!

We just open-sourced a robotics workload benchmark that tests compute platforms the way they’re actually used: running a full autonomous stack doing forklift pick-and-place, not simple isolated synthetic microbenchmarks. It processes 9 sensor streams from 3D LIDAR, depth cameras, etc running a full autonomy stack in a 180,000 sqft warehouse with a heavy VLM workload.

The benchmark is fully open-source, Dockerized, and easy to run on your own hardware. We’d love to see community results on other platforms. What compute platform is your team running? We benchmarked AMD Strix Halo, NVIDIA Jetson Thor and Orin AGX. Read about the key findings in our blog which also links to the full technical report!

Blog:

Want to add more elements of a full product or upload your results? Check out our GitHub page and plan to continue to support and maintain this for the long-term future! I think this can be a fabulous community resource to create a relatively detailed evaluation benchmark which grows over time with the needs of the industry - I know I’ll be using it when making decisions for future robot computer evaluations!

Happy benchmarking!

Steve Macenski

6 Likes

That looks like a very nice introduction of the AMD Kria AI platform that’s been unveiled today. It finally looks like a very good competition for the insanely expensive Jetsons.

Do you have any idea what price range will these kits be?

2 Likes

Thanks for comparison, nev Kria AI seems insanely powerful for such small device! Very funny they introduced it at the same time Nvidia raised prices by 50-100% :eyes: I’m also interested in price range.