Benchmark NPUs for an Autonomous Forklift Vision Stack
Overview
What this challenge is about.
You receive ONNX exports of the 3 production models, a labeled validation set of 2,000 forklift-camera frames, and developer-kit access to three NPU candidates (anonymized as NPU-A, NPU-B, NPU-C in the brief). Convert each model with each vendor SDK, benchmark end-to-end pipeline latency at 1080p, peak power draw, and accuracy retention vs. FP32. Score each NPU on a weighted matrix you justify (latency 40%, power 30%, accuracy 20%, SDK maturity 10%) and write a 3-page procurement memo with a clear recommendation and risks.
The Brief
What you'll do, and what you'll demonstrate.
Pick the right NPU for a 4-year autonomous-forklift program by benchmarking 3 chips on the actual perception workload and writing a defensible procurement memo.
Earning criteria — what you'll demonstrate
- Convert and deploy ONNX models across heterogeneous NPU SDKs
- Profile end-to-end pipeline latency, not just per-model latency
- Build a defensible vendor scoring matrix with sensitivity analysis
- Communicate procurement trade-offs to executive leadership
Program Fit
Where this fits in your program.
Sharpens the same skills your degree expects you to demonstrate.
Skills
Skills you'll demonstrate.
Each one shows up on your verified credential.
Careers
Roles this prepares you for.
Real titles. Real skill bridges. Pick the one closest to your trajectory.
AI Solutions Architect
Owning a multi-vendor NPU benchmark plus a defensible procurement memo is exactly the kind of high-stakes architecture decision AI solutions architects ship at scale-ups.
This challenge sharpens
- edge-inference
- npu-benchmarking
- vendor-evaluation
MLOps Engineer
Building a reproducible cross-SDK benchmark harness is core MLOps work that transfers to any team shipping ML to heterogeneous hardware.
This challenge sharpens
- onnx
- benchmarking
- model-deployment
Machine Learning Engineer
End-to-end pipeline profiling and accuracy retention checks across deployment targets is bread-and-butter MLE work at any edge-AI company.
This challenge sharpens
- edge-inference
- benchmarking
- onnx