Kris_KGoal is a new higher level cognitive robot architecture on top of ROS.; I'll post more about that soon (I have a bit up over on Researchgate). Mostly got this for research - want to compare it to the Xylinx Zinq ultrascale+ capabilities. (You get dual core RT on the Zinq, in addition to 4-core non-RT CPU plus the fabric; with the Xavier you get 8 cores with the 512 tesla core GPU. So I expect performance to differ depending on the algorithm, but mostly I'm interested in graph processing, not just ANNs... so will have to run my own benchmarks. Ultimate goal is a robot you don't program, but train as you might an assistant.
https://developer.nvidia.com/jetson-xavier https://developer.nvidia.com/embedded/jetson-xavier-faq https://www.cnet.com/news/nvidias-jetson-xavier-ai-chip-crams-10000-of-power-in-a-1299-cpu/
And they can slap two of those chips on a board for you with a couple extra GPUs. Price not yet announced, but just what you need for your autonomous air taxi project. I mean who isn't working on an autonomous air taxi?!
https://blogs.nvidia.com/blog/2018/01/07/drive-xavier-processor/