Senior Perception Engineer

Enable our humanoids to reliably detect the parts, fixtures and pallets they handle, and to recognize when they are uncertain. About Binabik AI Humanoid robots that run your machines. Binabik AI builds the software that lets humanoid robots do productive work on the factory floor. Our wheeled humanoids, AgiBot G2 and Galaxea R1 Pro, load and unload CNC lathes and mills, fetch material, carry parts and stack pallets. We focus on machine tending and intralogistics, the repetitive tasks that keep production running and are hard to staff. Our approach combines proven methods from classical robotics, such as motion planning, perception and control, with recent advances in physical and agentic AI. The classical parts make the system reliable and the AI parts make it adaptable. Teaching a new task takes only a few recorded demonstrations. Binabik Teach captures them, and our Skill Engine breaks the task into reusable skills. Binabik Sim, our custom simulator, then replicates each skill and refines it across variations in parts, positions and environment conditions before it goes to the real machine. A hardware abstraction layer lets skills carry over between robot platforms, and data collected in operation goes back into improving them. Customer data stays on-premise. We are a ZHAW spin-off with roots at the University of Zurich, the ETH AI Center and MIT. Our founders are experienced robotics and AI researchers who have published over 200 research papers. We are a pre-seed company finishing our first proofs of concept and moving towards industrial pilots. We plan to grow quickly over the next year, so the people who join now will help shape the product, the culture and the company. The role Every task our robots perform starts with perception. Whether it is a part in a bin, a chuck on a lathe or a partially loaded pallet, the robot must locate it precisely enough to grasp it, place it within tolerance and verify the result. As Senior Perception Engineer, you'll be responsible for how our humanoids perceive the shop floor, from raw camera and depth data to object poses, fixture localization and the scene representation used by our skills and planner. You'll combine classical geometric methods with modern learned models, including open-vocabulary detection, segmentation, 6-DoF pose estimation and vision-language models (VLMs). You'll make them run in real time on the robot under factory conditions such as reflective metal parts, clutter and changing lighting. This is a senior role: you'll define the technical direction for perception at Binabik.

What you'll do
• Own the perception stack. You'll be responsible for camera and hand-eye calibration, detection, pose estimation, tracking and the interfaces to skills and planning.
• Achieve the required precision. You'll localize parts, containers, fixtures and objects held in the gripper accurately enough for machine loading and palletizing.
• Deploy learned models on the robot. You'll fine-tune and deploy detection, segmentation and vision-language models with latency suitable for closed-loop control.
• Verify task outcomes. You'll implement visual checks that confirm whether each step succeeded.
• Ensure robustness. You'll improve depth quality on reflective and dark surfaces, handle occlusion and lighting changes, and estimate uncertainty so the robot can decide when to re-observe a scene.
• Bridge simulation and reality. You'll use Binabik Sim to generate training and test data, and validate on recorded real-world data before each release.
• Lead and mentor. You'll set the technical direction, review designs and develop perception into a reliable, well-tested subsystem. What you bring
• 5+ years of experience (industry, or PhD and industry) developing perception systems for real robots or other physical systems
• Strong 3D vision fundamentals: camera models, multi-view geometry, calibration, point clouds, pose estimation
• Hands-on experience training and deploying deep vision models with PyTorch or JAX
• Strong Python skills and experience optimizing GPU inference for latency
• Experience with ROS 2 and with debugging perception on live sensor data
• The ability to make and communicate technical trade-offs and to lead a subsystem, not only individual features Nice to have
• Open-vocabulary and foundation vision models (Grounding DINO, SAM, VLMs)
• Perception for manipulation: grasp detection and object pose tracking for industrial parts
• Synthetic data generation and domain randomization in simulation
• C++ and edge inference (TensorRT, ONNX)
• Publications at CVPR, ICCV, CoRL, ICRA or IROS The team Our engineers work on the whole pipeline from demonstration to deployed robot: perception, learned skills, simulation, the on-robot runtime and the tooling around it. Because the team is small, research and product work happen side by side. The person who trains a skill is often there when it loads a real lathe for the first time. How we work
• Real hardware, every week. Everything we build is meant to run on our G2 and R1 Pro in the lab, and from there on a customer’s shop floor.
• We build the robot software and the tools to develop it. That includes the on-robot stack, the simulator and the teaching tools. When a tool slows us down, we can fix it ourselves.
• Out of the lab early. We put new capabilities in front of customers quickly and evaluate them in real production settings. Engineers join these deployments, and what they see there shapes what we build next.
• AI-assisted development. We use coding agents and LLM tools every day, and we care

about us
ing them well. Our stack: Python (backend and most of the robot stack), ROS 2, MuJoCo, PyTorch, GPU compute in the cloud, and React + TypeScript for frontends.

What we offer
• Compensation and early influence: a competitive compensation package. As one of the first employees, you will help shape the product, the company’s direction and how we work.
• Hardware access: You’ll work hands-on with our humanoid robots, lab equipment and GPU compute from your first day.
• Career development: Your role can change as you do. Whether you want to specialize technically, take on more responsibility or move into a new area, we will support it.
• Customer and public exposure: You’ll take part in customer visits and deployments, and you can represent Binabik at industry events and conferences.
• Team: You’ll join colleagues with backgrounds in robotics and AI research who enjoy solving difficult problems and bringing them into practice.
•

Benefits:
a pension plan well above the legal minimum, comprehensive insurance coverage and 5 weeks of paid holiday.
•

Location:
Our office at Technopark Winterthur, with the robot lab on site, is a short train ride from Zürich. Hiring process
• Introductory call: a short conversation about your background, your goals and what we’re building.
• Technical interview: We go into your experience in depth and work through a problem drawn from our day-to-day work. For commercial roles, this step is a case discussion.
• A day with us: You’ll spend a day at our office in Winterthur, see the robots in operation and meet the team.

How to apply
Please fill out the form on Dover. If you like, include a link to relevant work, such as a repository, a paper, a robot video or a portfolio. If you don't meet every requirement, we still encourage you to apply. We value curiosity and potential as much as a complete match with the list. Our working language is English. German is a plus for customer-facing roles. We can currently consider applicants who are EU/EFTA citizens or who already hold a Swiss work permit. Binabik AI is an equal opportunity employer. We welcome applicants of every background, gender, age, origin, religion, sexual orientation and disability.