Munich Autonomous Driving Meetup

I am honored to have opportunity to speak at this distinguished gathering alongside experts in AI and industry

Munich Autonomous Driving Meetup

Two announcements: 1) Feb 24 xLSTM Evening w/ Sepp Hochreiter, NXAI, Siemens DI SW and Spleenlab! https://lnkd.in/eyQC_772 2) Out Meetup series extends to Autonomous Driving, Robots & AI!

As you can see below, there was no other choice but to set up a small fireworks particle simulator with a dancing robot to celebrate!

___ Agenda ___

_ Deep Dive into xLSTMs Prof. Sepp Hochreiter, JKU Linz

_ Building a company on xLSTM: From Cutting-Edge Research to Industrial Applications Albert Ortig, CEO NXAI

_ AI-Driven ADAS Validation in the Software-Defined Vehicle Era Gwen van Vugt & Son Tong, Siemens Digital Industries Software

_ xLSTM for Physical AI: Opportunities and Challenges for Vision based Robotics from drones to cars Victor Kallenbach, Stefan Milz, Spleenlab

___ Further Details ___ _ If you attend the Meetup, you give declaration of consent that pictures that may show you may be taken and may be used in for publications by the event host or meetup organizer group for public promotion or review of the individual events or event series, on social media or other media. May! _ The talks are all held in English and will take ~15-25 minutes each + ~10 minutes Q&A

  • NXAI is a EU based startup based on the xLSTM technology.
  • First phase was to demonstrate capability of xLSTM - training 7B model, benchmark to prove efficiency and scalability
  • xLSTM models can be increased in size and are expected to scale well, other means of optimization like Mixture of Experts can be applied; but NXAI focus is on embedded and industrial applications with small models.

_ AI-Driven ADAS Validation in the Software-Defined Vehicle Era Gwen van Vugt & Son Tong, Siemens Digital Industries Software

  • With increasingly complex ADAS/AV functions, validation effort explodes
  • Siemens solves via SiL and MiL, key is to make Sim scalable and accurate
  • Finding Unknown Unsafe scenarios is key - instead of focussing on individual actors, Siemens approach is to have other actors collaborate in order to provoke unsafe scenarios for ego vehicle - while respecting physics
  • Real2Sim/Prescan Simulator: leveraging Gaussian Splatting and GenAI for realistic environments & scenarios

_ xLSTM for Physical AI: Opportunities and Challenges for Vision based Robotics from drones to cars Victor Kallenbach, Stefan Milz, Spleenlab

  • Spleenlab builds embedded AI solutions for drones cars, … e.g. Drone navigation in GPS denied environments, depth reconstruction from fisheye cameras, etc.
  • xLSTM promises efficiency for compute-constrained embedded platforms -> ran experiments
  • CUT3R for 3D scene reconstruction is great but too slow for embedded targets -> replace transformer by xLSTM to improve throughput
  • benchmarking Vision xLSTM against DiT on Kitti Eigen Benchmark -> RMSE on par without further optimization

Thanks for joining, special thanks to Robert Weber & Boris Scharinger for co-organizing, Siemens DI & NXAI for hosting us, and Florian Netter helping to run the Meetup overall!

Speakers: please share links to your slides in comments, if possible!

Looking forward to the next Meetup - now with broadened scope: Autonomous Driving, Robots & AI!