Autoware on ARM Improvements

on YouTube ↗17m

" summary

Code coverage rises from 0.7% to 2.7% lines and 1.1% to 4.5% functions after adding unit tests and gating PRs; CI moves from Travis to GitLab CI via mirrored GitHub repos but faces fork-trigger and x86-runner limits, prompting evaluation of Jenkins and on-prem ARM64 runners. Dockerfiles unify cross-arch builds with a CUDA-enabled image and separate arm64/x86 repos. CUDA dependency is removed for YOLO v3 in Autoware 1.11; preliminary integration tests run with an on-device gyrfalcon accelerator and ARM NN back-end, plus multithreading fixes and benchmarking across CPU/GPU configurations.

AI-written summary · about

" chapters (9)

1. 0:00Introduction and overview
2. 1:09Code quality and testing improvements
3. 2:52CI migration and pipeline changes
4. 5:45Documentation and Docker updates
5. 8:58Object detection module changes
6. 10:29ARM NN and inference engine
7. 13:22Multi-threading and performance
8. 14:46ROS 2 migration plans
9. 15:52Summary and resources

" related talks

• 20m96Boards - Autoware: 1 year onServando German Serrano · Linaro Tech Days 2020
• 28mOpenAdx "Autonomous Driving Simulation" Testbed – hardware in the loopMark Walton · EclipseCon Europe 2018
• 01mPhysical AI is everywhereDrew Henry · CES 2026
• 29mTeam Teardown: MLOps in the Cloud for IoTDanielle Dean, Mathew Salvaris, Mohan Mupiddi · TWIMLcon: AI Platforms 2021
• 47mAchieving Software Quality Excellence Thru Software-Hardware Co-Design & Co-VerificationMonica Bao · PNSQC 2024
• 30mBreaking Heterogeneity Barriers: Unified Cloud-to-Robot AI System SW Stack for...Yonghua Lin · PyTorch Conference 2025
• 31mDeep learning for self-driving vehicle technologyInmar Givoni · Big Data & AI Toronto 2020
• 24mAI in maritime - Autonomous shipsMarina Ivasic Kos · DSC Europe 24