Confirmed on the employer's own hiring board on Nov 17, 2026. First seen by Alion on Nov 17, 2026.
Analog is building robots that work in the real world for governments and enterprises, and they’re not just in demos. They’re deployed, and tested on-site in Abu Dhabi. Right now, a robot’s ability to understand what’s in front of it is one of the biggest things standing between a good robot and a useful one. You’d be the person who owns that.
It’s a rare moment to join. We’re a 80-person team, in one city, building something that doesn’t exist anywhere else yet, and perception is a problem we’re still shaping rather than inheriting.
What we’re looking for, and why
Someone who has already shipped vision and perception on real robots and knows the gap between a model that works in a notebook and one that works on a small machine with a tight power budget.
What you’ll be working on
Helping our robots detect and understand people, objects and the space around them, using RGB, depth and LiDAR
Turning what a camera sees into where things actually are, in the robot’s frame and on the map
Understanding relationships in a scene, not just detecting things. Think two people talking while looking at a TV, or “pick up that cup”
Getting models fast and small enough to run in real time on the robot, with no waiting on the cloud
What makes this role different
It’s real hardware. Your work runs on robots in the building, so you see it succeed or fail the same day.
Small robots, real constraints. We care about humanoids and small mobile or legged robots, where battery and compute are tight. Optimisation is the job, not an afterthought.
Flat and fast. No layers between you and the people building the robots, so your decisions ship.
Perception is the foundation. Everything our robots do starts with what they can see.
Skills we need
5+ years in computer vision or perception for robots or autonomous vehicles, deployed on real hardware
Detection of people, objects and other entities on multi-modal data (RGB, RGB-D, LiDAR)
Grounding detections from 2D images to 3D positions in the robot and map frames
Multi-sensor fusion of camera, depth and LiDAR, plus tracking and re-identification
Vision-language models: grounding language to entities in the scene
Model optimisation for real-time use on edge GPUs: TensorRT, quantisation, pruning, CUDA
ROS 2, C++, Python, PyTorch
Nice to have
Open-vocabulary detection and segmentation
Camera and LiDAR intrinsic / extrinsic calibration
Synthetic data and simulation for training and testing (e.g. Isaac Sim)
Human pose estimation, gaze and intent for human-robot interaction
What we’re not looking for
People whose experience is only cloud-based models with no real-time or on-device work
People who want to manage a team rather than build
People who need a lot of process and structure before they can move
People who want a remote setup. We’re in person, in Abu Dhabi.

