Staff Perception Engineer – Robotics

About the role

The role involves owning the full perception stack for Knightscope's autonomous security robot platform, including raw sensor data through 3D object detection, scene understanding, and persistent entity tracking. Responsibilities include designing, training, and deploying 3D perception models, managing the ML pipeline, and implementing evidence-first capture.

Responsibilities

• Design, train, and deploy 3D perception models, object detection, segmentation, and multi-object tracking, from multi-modal sensor data (LiDAR, camera, thermal, radar) using state-of-the-art BEV and transformer-based architectures.
• Own the end-to-end ML pipeline from large-scale data curation and annotation strategy through model training, optimization (TensorRT/CUDA), and real-time onboard deployment on edge compute.
• Build persistent cross-sensor identity: the same person, vehicle, or license plate maintains a stable track across camera handoffs, patrol legs, and time, feeding entity correlation and incident generation downstream.
• Design and implement evidence-first capture: pre/post-event clips, full-frame and crop media, and structured detection output that is replay-safe and analyst-ready.
• Explore and integrate Vision-Language Models (VLMs) to enrich detection outputs with scene narration, anomaly reasoning, and long-tail incident understanding, moving K7 beyond bounding boxes toward analyst-ready scene descriptions.
• Drive technical decisions on architecture, data strategy, and roadmap; mentor engineers across the team.
• Design and run rigorous offline and closed-loop evaluations; define metrics for safety-critical perception performance across diverse real-world deployment environments.

Requirements

• 7+ years of hands-on experience at an OEM, AV company, physical security platform, or robotics tech company shipping perception software to production.
• Deep expertise in 3D computer vision: LiDAR/camera 3D object detection, point cloud processing, multi-view geometry, and sensor fusion.
• Strong ML fundamentals, CNNs, Transformers, multi-task architectures, with production deployment experience (TensorRT, ONNX, CUDA optimization).
• Fluency in C++ (real-time systems) and Python; experience with ROS2 and simulation environments (CARLA, Isaac).
• Production-first mindset demonstrated ability to take models from research to deployed, real-world systems operating under real operational constraints.

Preferred Qualifications

• Exposure to vision-language models (VLMs): fine-tuning, prompting, or integrating VLM outputs into a perception pipeline for scene understanding or anomaly narration.
• Familiarity with Vision-Language-Action (VLA) or end-to-end policy models that map sensor observations directly to robot actions, and interest in applying these to active confirmation and repositioning behaviors.
• Experience with ALPR, face recognition, or ReID systems in deployed security or automotive contexts.
• Familiarity with evidence integrity, chain-of-custody media, or tamper-evident capture pipelines.
• Publications or open-source contributions in 3D CV, embodied AI, or autonomous systems.
• MS or PhD in Computer Science, Robotics, Electrical Engineering, or related field.

Compensation & Benefits

Base Salary: $240,000 to $275,000 (DOE)
Equity: Stock options
Benefits: Medical, dental, vision, 401(k), paid time off

Location Requirement

Full-time, on-site at Sunnyvale HQ (No relocation provided)