Perception Engineer - Offroad

Applied Intuition•Sunnyvale, CA

1d•$180,000 - $255,000•Onsite

About The Position

As a perception engineer on the Mining Autonomy team, you will own perception outcomes for autonomous mining vehicles operating across active customer sites in Australia, Europe, and the US. You are responsible for making our core autonomy platform work in the real world: selecting sensor sets for new vehicles, adapting existing models to new domains and operational design domains, diagnosing why a model underperforms on a new site, and deciding when to fix it yourself versus routing it back to the core team. You will work at the intersection of perception, planning, controls, and customer requirements, translating what a mining customer needs into a concrete technical approach and a set of KPIs the team can build toward. You will collaborate daily with the core autonomy platform team, customer engineering, and BD, and will help stand up new product lines alongside developing existing ones.

Requirements

Strong systems-level understanding of perception (and ideally planning/controls) — enough to reason about the full stack, even without owning model training personally
A product-owner mindset: motivated by customer success and problem ownership more than by being on the cutting edge of modeling research
Experience diagnosing real-world model failures in deployed systems and driving them to resolution, whether hands-on or through cross-team negotiation
Comfort working with ambiguous, evolving requirements from customers and translating them into technical specs
Familiarity with PyTorch, ONNX, TensorRT, Docker, Python, some C++

Nice To Haves

Experience with internal ML pipeline/orchestration tooling (e.g., Flyte) is a plus
Experience with offroad autonomy

Responsibilities

Design and implement perception systems to enable new product capabilities, leveraging existing model heads and solving new perception tasks.
Design the data strategy for a new deployment: what to collect, how much, and how to label it, to get a model production-ready
Make build-vs-reuse tradeoffs when standing up new product lines, including output types, KPIs, and collaboration points with the core perception team
Triage model and system issues in the field — timestamps, data quality, calibration, and similar — and decide whether to resolve them directly or work with the core platform team to fix them upstream