Perception Lead

nerostechnologies; Torrance (CA); On-site; USD 180,000 - 240,000; Full time

Job summary

Neros seeks a Perception Lead to own the perception stack across its unmanned aircraft. You will lead a small team and contribute directly as an engineer, delivering robust detection, recognition, and tracking from sensor inputs to state estimation.

You will steer transitions from classic computer vision to learned methods, while ensuring latency budgets and onboard compute constraints are met, including deployment to embedded accelerators.

Qualifications

• 6+ years developing production computer vision or perception software.
• 2+ years in tech lead or engineering manager role with hands-on work.
• Delivery of perception software to physical systems in outdoor environments.
• Expert proficiency in C++ and Python for real-time or resource-constrained systems.
• Experience deploying neural networks to embedded accelerators (Jetson, TensorRT, etc.).
• Knowledge of EO/IR imaging, camera calibration, and long-range small-object detection.
• Ability to define perception evaluation metrics and regression testing.
• Experience setting detection thresholds against false positives and misses.

Responsibilities

• Own perception stack architecture, roadmap, and delivery across vehicle platforms.
• Lead and grow the perception engineering team, while contributing as an engineer.
• Deliver detection, recognition, and image-space tracking under degraded conditions.
• Define perception output contracts with calibrated confidence signals.
• Advance from classical CV to learned detection, with deterministic fallbacks.
• Deploy perception software to constrained onboard compute with latency considerations.

Skills

Computer vision

Perception software

Team leadership

Field deployment

C++

Python

Embedded NN deployment

IR imaging

Performance evaluation

Sensor timing

Sensor fusion

Tools

NVIDIA Jetson

TensorRT

Qualcomm

Hailo

Job description

Who we are

Neros is a defense technology company rebuilding America's drone industrial base. We design and manufacture high-performance unmanned systems that are tested in combat, iterated at startup speed, and built at massive scale. Our team culture is fast, hands-on, and obsessed with closing the gap between design and deployment.

As drones transform the character of warfare, Neros is delivering the systems the West needs to compete on the modern battlefield and deter the adversaries of democracy. We're hiring engineers, operators, and builders who want to move fast, take on extreme ownership, and get capability into the hands of warfighters in months, not years.

What you will be doing

• The Perception Lead owns the perception stack across all Neros unmanned aircraft, covering detection, recognition, and image-space target tracking from the sensor through to the target track consumed by state estimation and guidance. This role leads a small perception engineering team while contributing directly to the software, and sets the technical direction for the transition from classical computer vision to learned models as flight data justifies it. The Perception Lead is accountable for perception performance under operationally degraded conditions, including long standoff, low light and thermal imaging, cluttered land and maritime backgrounds, occlusion, and loss of communications.
• Own the architecture, roadmap, and delivery of the perception stack across all vehicle platforms, maintaining a single configurable implementation rather than per-program forks
• Lead and grow the perception engineering team through hiring, mentorship, and technical review, while continuing to contribute directly as an engineer
• Deliver detection, recognition, and image-space tracking that hold a designated target and re-acquire it after loss, in clutter, glare, low light, occlusion, and degraded link conditions
• Define the perception output contract, including calibrated confidence and explicit signaling of suppressed or degraded detection, and set detection operating points jointly with the state estimation and engagement authorization owners
• Drive the progression from classical computer vision to learned detection and classification, determining when flight data demonstrates that a learned component outperforms the deployed baseline, defining the evaluation criteria and regression cases that decision rests on, and retaining a deterministic fallback
• Own deployment of perception software to constrained onboard compute, including accuracy and latency trade-offs, optimization for embedded accelerators, and compliance with end-to-end latency budgets

You should have the following

• 6+ years developing production computer vision or perception software, including 2+ years leading a team or a technical program as a tech lead or engineering manager, with continued hands-on technical contribution
• Demonstrated delivery of perception software onto a physical system operating in an uncontrolled outdoor environment, such as unmanned aircraft, ground robotics, autonomous vehicles, or aerospace platforms
• Expert proficiency in C++ and Python, including production experience under real-time or resource-constrained requirements
• Depth in both classical and learned perception, spanning correlation, template and optical-flow tracking, feature matching and data association, alongside modern detection and classification networks
• Experience deploying neural networks to embedded accelerators such as NVIDIA Jetson and TensorRT, Qualcomm, Hailo, or equivalent, including quantization and characterization of the resulting accuracy and latency trade-off
• Working knowledge of EO and IR/thermal imaging, camera calibration, and sensor characteristics, with experience in small-object detection at long range, in low light, or under high dynamic range
• Experience defining perception evaluation, including dataset and labeling requirements, detection and tracking metrics such as precision, recall, track continuity and identity switches, and regression testing against logged flight or field data
• Experience selecting detection thresholds and operating points against an explicit cost of false positives versus missed detections in a safety-critical or mission-critical system
• Familiarity with sensor timing and synchronization, coordinate frames, and uncertainty representation at the interface to state estimation and navigation

Nice to Have

• Experience selecting and integrating thermal/LWIR sensors, including the imaging trade-offs and procurement constraints that come with them
• Experience building or operating a data engine for a fleet, covering large-scale log capture, indexing and search, dataset versioning, and labeling operations that feed model training
• Working familiarity with state estimation and sensor fusion, such as visual-inertial odometry or Kalman filtering, sufficient to reason fluently across the boundary between perception and navigation
• Experien