Computer Vision Engineer — L3/L4
PerceptyneNot specifiedOn-Site
Needs a free Superjobs account. You'll finish on Perceptyne's site.
Skill score
Summary
This person builds and maintains real-time computer vision pipelines that let Perceptyne's industrial humanoid robots perceive parts and workspaces for grasping, insertion, and inspection, validating everything on physical hardware. It suits a senior, hands-on engineer with 5+ years of production CV experience who combines traditional geometric methods with modern deep learning and can debug failures across optics, sensors, and robot control.
- Function
- Development
- Estimated salary
- Est. ₹11–16 LPA
- Estimated for a mid-level role, ~5 yrs experience, India. Not published by the employer.
- Work mode and location
- On-Site, Not specified
- Experience
- 5+ yrs
Job description
Overview
Hyderabad, India · Full-time · On-site with real robots
In a factory, vision is rarely clean: parts reflect light, cables occlude features, objects arrive in unexpected orientations, components vary between batches, and a few millimetres of pose error can turn a successful insertion into a failure. At Perceptyne, computer vision is not a separate analytics function—it is part of the real-time system that lets a robot understand its workspace and use its hands.
If you want to be a part of building the perception stack and build vision systems that directly determine whether a physical robot can grasp, align, insert, inspect, and recover from errors in the real world, this is that job.
- This is a senior hands-on role for someone who can independently turn an ambiguous perception problem into a reliable system running on a robot.
Roles and responsibilities
- You will own significant parts of the perception stack from raw sensor data to robot action. You will work closely with manipulation, controls, mechanical, electronics, and AI engineers—and validate your work on physical robot hardware.
- Design and deploy robust, real-time perception pipelines for manipulation and assembly tasks.
- Build deployment grade modular computer-vision code using / improving upon existing libraries for calibration, feature extraction, geometric matching, image enhancement, segmentation, tracking, metrology, and inspection.
- Develop methods for 2D/3D object detection, pose estimation, registration, grasp localization, alignment, and visual servoing.
- Work with RGB, stereo, depth, and industrial cameras; own practical aspects of imaging such as selection, camera placement, triggering, synchronization, exposure, lighting, calibration, and failure detection.
- Build sensor-fusion pipelines combining cameras, depth sensors, force / torque sensors, touch sensors robot kinematics, encoders, and other relevant signals.
- Implement robust image-processing techniques for difficult real-world scenes: glare, reflections, low texture, motion blur, occlusion, changing illumination, transparent or reflective surfaces, and part-to-part variation.
- Design data-collection and evaluation workflows, including ground-truth capture, dataset curation, test-case construction, metrics, error analysis, and regression testing.
- Integrate classical CV, geometric reasoning, and learned models where each is appropriate. Use deep learning when it improves the system
- Optimize pipelines for edge deployment, including latency, throughput, memory use, GPU/CPU utilization, and reliable production operation.
- Prototype in offline datasets and simulation, then validate results on real robots. The robot and the production environment are the final benchmark.
- Debug cross-functional failures where the root cause may be in optics, illumination, synchronization, calibration, sensor drivers, robot kinematics, coordinate transforms, inference, or a downstream planner.
Skills required
Top skills
- C++
- Python
- OpenCV
- Traditional computer vision
All skills (12)
- Camera calibration
- Stereo geometry
- Multi-view geometry
- Pose estimation
- Edge deployment
- Linux
- Industrial cameras
- Sensor fusion
- PCL or Open3D
- ROS 2
- Hand-eye calibration
- TensorRT
Qualifications
- Bachelor’s, Master’s, or PhD in Computer Science, Electrical/Electronics, Mechanical Engineering, Robotics, Computer Vision, or an equivalent technical field.
- 5+ years of hands-on computer-vision engineering experience, including technical ownership of production or field-deployed vision systems.
- Strong C++ and Python skills. You should be comfortable writing performant production code, debugging native libraries, profiling bottlenecks, and working in Linux-based environments.
- Deep practical experience with traditional computer vision and image processing—not only deep-learning-based techniques.
- Strong working knowledge of OpenCV or equivalent production-grade CV libraries.
- Experience with techniques such as camera calibration, stereo geometry, image rectification, filtering, thresholding, morphology, edge and contour extraction, feature extraction/matching, template matching, optical flow, geometric transforms, tracking, image registration, and image-quality evaluation.
- Solid understanding of multi-view geometry, projective geometry, coordinate systems, transformations, camera models, and 3D reconstruction or pose estimation.
- Demonstrated ability to diagnose why a vision pipeline fails in real images and improve it through optics, lighting, data, algorithm design, calibration, or system integration.
- Experience deploying CV pipelines on edge compute, with attention to real-time latency, frame rate, determinism, reliability, and resource limits.
- Experience in robotics, industrial automation, machine vision, manufacturing inspection, bin picking, or assembly automation.
- Experience with industrial cameras and standards such as GigE Vision, USB3 Vision, synchronization and multi-camera systems.
- Experience with depth cameras, stereo vision, structured light, time-of-flight sensors, LiDAR, or point-cloud processing
- Grit and dedication: you stay with difficult, ambiguous technical problems through repeated failed experiments and work systematically until the system is robust—not merely successful in one controlled demo.
Nice to have
These are valuable but not mandatory. Apply if you are strong in the core requirements even if you do not match every item below.
- Familiarity with PCL, Open3D, NVIDIA VPI, CUDA, models on Omniverse , TensorRT, or GStreamer
- Experience with robot-camera hand–eye calibration, visual servoing, eye-in-hand systems, or vision-guided manipulation.
- Knowledge of ROS/ROS 2, TF/coordinate-frame management, MoveIt, rosbag, RViz, Gazebo, Isaac Sim, or robot simulation.
- Experience with machine learning for perception, including CNNs, transformers, segmentation, representation learning, synthetic-data generation, or model deployment.
- Experience designing fixtures, lighting, optical setups, or test rigs for machine-vision systems.
- Contributions to relevant open-source projects, technical publications, deployed products, patents, or a portfolio of working CV systems.
Other details
Modern AI is very valuable, but it is not the whole perception stack.
For some robotics tasks, the system needs precise geometry, explainable failure modes, reliable timing, and stable behaviour even when the available dataset is small or the part geometry changes.
We are specifically looking for engineers who can build and troubleshoot that full stack that integrates both Modern AI and traditional CV based methods.
You should be comfortable selecting an approach based on the problem:
Company info
About Perceptyne
Perceptyne is a deep-tech robotics company building AI-powered semi-humanoid robots that automate dexterous and intelligent tasks on industrial shopfloors. Its robots are used in areas such as electronics product assembly, automotive assembly lines, and warehouse operations.