Computer Vision Engineer
Revel Robotics · Prague ·
- Work mode
- Onsite
- Seniority
- Intern
- Employment
- Full time
- Category
- ML ai
- Experience
- 3+ years
Revel Robotics · Prague ·
bedrock-robotics · San Francisco, CA
trener-robotics · Trondheim, Norway
Origin · San Francisco, California, United States
Clausal · Helsinki, Helsinki, Finland
Computer vision engineer at an AI robotics startup building the perception stack for the GENESIS humanoid robot in Prague. Day to day: 3D hand/tool/object tracking from multi-camera and wearable data, detection/segmentation/pose estimation, camera calibration, and deploying real-time models on the robot's GPU with Python, PyTorch, ONNX/TensorRT.
ABOUT REVEL
REVEL is an AI robotics company developing physical intelligence for general-purpose humanoid robots. We capture the force, dexterity, and intent of human work with our Neural Gambit wearable, and use it to train RAI, the intelligence that powers our robots. REVEL is headquartered in Palo Alto, California, with R&D and engineering facilities in Prague and Hradec Králové, Czech Republic. This role is on-site in our Prague lab.
THE ROLE
We are looking for an experienced computer vision engineer to build the perception that lets our humanoid robot, GENESIS, see and understand its workspace in real time. You will turn REVEL's recordings of skilled human work, captured with several cameras and the Neural Gambit wearable, into models that track hands, tools, and objects in 3D. Then you will get those models running on the robot.
Every manipulation skill the robot learns depends on seeing hands, tools, and objects accurately. Perception errors show up as failed grasps on a customer's floor.
You will join a small Perception team in Prague and work closely with our Robot Learning and Data Collection teams. Most of the team is early in their careers, so you will help set how we build, evaluate, and ship models, and support the junior engineers along the way.
WHAT YOU'LL DO
Build hand, tool and object tracking in 3D from our multi-camera and wearable recordings of people doing manual work.
Develop detection, segmentation, pose estimation and 3D scene understanding for the robot's manipulation skills.
Calibrate and synchronise multi-camera and depth setups in the lab and on the robot, and keep them accurate over time.
Build training and evaluation pipelines that anyone on the team can rerun months later: data, labelling, dataset versioning and experiment tracking.
Design evaluations around what the robot needs, not only benchmark scores, and find out why a model fails on real data.
Get models running in real time on the robot's onboard GPU (ONNX, TensorRT, quantisation or similar).
Work with Robot Learning and Data Collection on what to record and what perception the next skill needs.
Help set engineering standards in a small, young team.
WHO WE'RE LOOKING FOR
3+ years of hands-on computer vision work in industry or research, with models you trained yourself running on real camera data, not only on benchmarks.
Strong Python and PyTorch.
Solid Linear Algebra and 3D Computer Vision fundamentals.
Deep-learning fundamentals for vision (detection, segmentation, pose estimation or 3D vision) and rigorous evaluation.
Experience with real cameras and real-world data: calibration, multi-camera setups, noise, lighting and latency.
Experience taking a model from training to real-time inference on a robot or an edge device.
Good engineering practice: Git, Linux, clean code and reproducible experiments.
Honesty about what you know. You say "I don't know" when you don't, take criticism of your ideas well, and still defend a good idea with evidence.
Fluent English for daily work with colleagues in Prague, Hradec Králové and the US.
Able to work full-time on-site at our Prague lab.
BONUS POINTS, BUT NOT ESSENTIAL
Hand and body pose tracking, or egocentric video.
Depth sensors, point clouds or 3D reconstruction.
Experience with IMU and sensor fusion.
Visual SLAM, VGGT and any other modern NN-based methods in computer vision.
Hands-on hardware: you have integrated sensors on a real robot or product prototype.
Image processing pipeline: from raw sensor data to .jpg.
ROS 2.
OpenCV, numpy, scipy, pandas/Polars.
Vision-language-action or foundation models.
Publications in computer vision, ML or robotics.
WHAT THIS ROLE IS NOT
This is not a research-only role. Success means models that work on the robot, and getting them there is part of your job.
This is not a benchmark role. You will work with real cameras, real lighting and messy recordings every day.
This is not a junior or internship position. We are looking for someone who can own a perception problem from the data to the robot.
WHAT YOU'LL GET OUT OF IT
The chance to build the eyes of a humanoid robot, trained on REVEL's own recordings of skilled human work.
Data most computer vision engineers never get: multi-camera recordings of real manual work, together with what the Neural Gambit wearable captures.
Your models run on robots in the same lab where you work.
Real ownership of the perception stack, and room to take on broader responsibility as the team grows.
WHY REVEL
The Team: Colleagues from NVIDIA, SpaceX and Neura Robotics, in a small team with short paths to decisions.
Real Technology: You see the robots in the lab every day, not just in a pitch deck.
A Fast-Growing US Startup: Founded in 2025 and building its team in the Czech Republic.
WHAT WE OFFER
Competitive compensation based on experience.
Unlimited Paid Time Off: Real flexibility to take time away when you need it. We trust our people to own their work, their time and their results.
Lunch, On Us: Complimentary lunch every working day, plus coffee, snacks and drinks whenever you need a boost.
Prague, On-Site: Our office is at U Pekáren 1645/1, Prague, Hostivař.
HOW TO APPLY
Apply through the link on this posting. Along with your CV, you will fill in a short form.