VYXA — Computer Vision Engineer

About VYXA

VYXA is an early-stage physical AI company building consumer robots and real-time vision systems that work in the messy, dynamic real world.

Our products combine custom hardware, on-device intelligence and a software agent. Perception sits at the core: the system must see, understand and respond to fast-moving people and objects, outdoors, in real time.

We are a venture-backed team in Sydney, roughly half software and half hardware, with an ambitious hardware roadmap for 2027.

The role

You will be our first dedicated computer vision hire and own the perception stack end to end: from raw camera frames to the signals our robots and software act on.

Your work spans two surfaces: a mobile robot with onboard cameras and an actuated camera head, and fixed multi-camera installations. Both feed one shared model of the people, objects and environment around them.

You will work directly with the CEO, the software co-founder and the hardware lead. This is a full-time, on-site role in Greater Sydney.

What you'll do

- Small, fast-object detection and tracking. Detect and track small, rapid-moving objects in real time under harsh outdoor light, hard shadows, occlusion and cluttered backgrounds, on mobile and embedded hardware.
- Person tracking. Build person detection, re-identification and tracking that holds the right target through occlusion and crowded scenes, and drives active camera control and follow behaviour.
- Scene geometry. Detect structure and keypoints, calibrate cameras and estimate homography, and map people and objects into world coordinates for navigation and planning.
- Trajectory and event estimation. Estimate 3D trajectories and contact events, fuse multiple camera views, and deliver high-accuracy real-time decisions.
- Pose and action recognition. Use human pose and action recognition to power activity analysis and feedback.
- Edge deployment. Quantise,



prune and optimise models to hit latency and power budgets on smartphones (Core ML, TFLite) and embedded compute.
- Data engine. Build capture, labelling and evaluation pipelines from our own real-world footage, define the metrics that matter, and use synthetic data where it helps.
- Field testing with hardware. Work with the hardware team on camera choice, placement, field of view and exposure, and test in real environments every week.

What we're looking for

- PhD in computer vision, robotics, machine learning or a related field, or an MSc with an equivalent research and shipping record.
- Deep experience in object detection and multi-object tracking, ideally for small or fast-moving targets.
- Strong Python and PyTorch; working C++ for performance-critical inference.
- Solid multi-view geometry: camera calibration, homography, triangulation and 3D reconstruction.
- A track record of taking models from research to real-time deployment on edge or mobile hardware.
- Comfort with ambiguity, fast iteration and testing outdoors, in a team where you set direction for your area.
- Right to work in Australia and able to work on-site in Sydney.

Nice to have

- High-speed object tracking experience, in sports analytics, broadcast, industrial inspection or similar.
- Robotics perception: visual servoing, SLAM, sensor fusion, ROS 2.
- On-device ML on iOS or Android.
- Human pose estimation and action recognition.
- Publications at CVPR, ICCV, ECCV, NeurIPS, ICRA or IROS.
- Mandarin, for working with our manufacturing partners.

Why join

- Founding-team ownership. You define the perception roadmap and the architecture, not a slice of someone else's.
- Real-world impact, fast. Your models ship in hardware on a 2027 launch roadmap, in front of real users.
- Hard problems. Small fast objects, outdoor light, edge compute and multi-camera fusion, all in one system.
- Compensation. Competitive salary plus meaningful early-stage equity.

How to apply

Send your CV and one project you're proud of (a paper, repo or demo video) and a few lines on why it was hard.

📌 Computer Vision Engineer (Australia)
🏢 VYXA
📍 Australia