Writing Safety Critical Automotive Software for High Perf AI Hardware
summary
Summary of concrete claims and work: automotive ML and sensor stacks produce 120 TB/month of sensor data and industry uses a billion‑mile metric; analyses cite real crashes where camera/radar disagreement or intentionally disabled systems caused fatalities. Standards work converges around MISRA as the base, AUTOSAR material folded into it, and SG‑12/WG23 feed concrete, automatable rules back into C++ core guidelines. Roadmap targets a safety‑critical SYCL and explicit rules for concurrency and heterogeneous dispatch so AI accelerators can be certified under ISO 26262.
AI-written summary · about
chapters (18)
- 0:00Introduction
- 5:00Recap
- 6:10What does it take to drive
- 7:50ISO 26262
- 9:20Selfdriving car crashes
- 12:35Tesla autopilot crash
- 13:47Uber autonomous vehicle crash
- 16:38AI accelerators
- 19:26Safety Critical Systems
- 26:17C Guidelines
- 30:47C Safety
- 36:02Comparing coding standards
- 37:17Memory management
- 38:47Memory leaks
- 39:47Minimize fragmentation
- 43:12Misra
- 46:12Why am I here
- 48:22What Im trying to do
related talks
08mHuman–AI Collab for Reliable Automotive Control Systems
24mUse of AI to quantify the "safety of autonomous vehicles"
08mEmbedded Executive LIVE
08mdata science in automotive
19mBoosting confidence in tiny AI deployment in embedded control systems
18mAutomated Risk Evaluation and Mitigation for AI Systems
09mLightning Talk: An Efficient Automotive Safety Support System Utilizing Hardware...
02mAI-defined vehicles and autonomy