Our pick-and-place prototype is running well mechanically, but its accuracy now depends on precise vision alignment. An Intel RealSense RGB-D camera is mounted on the robot’s wrist, and I need robust eye-in-hand calibration to lock its coordinate frame to the manipulator’s. The existing pipeline is written in Python with heavy use of OpenCV; your work will slot directly into that codebase. The focus is: • Generate reliable camera-to-robot extrinsics through established hand-eye methods (e.g., Tsai–Lenz, dual-quaternion, or comparable approaches). • Verify the resulting transformation live on the robot and diagnose any residual frame drift or axis flips. • Feed the calibrated poses back into our grasp-planning module so picks land within ±1 mm repeatability. All development is remote; a VPN link gives you shell access to the robot PC and a stream from the RealSense. Regular screen-share test sessions will be scheduled during IST working hours, so being based in India is ideal but not mandatory. Deliverables 1. Well-documented Python calibration script callable from our main repo. 2. A brief README detailing setup steps and expected accuracy benchmarks. 3. Short report summarising test results and any recommendations for maintaining calibration over time. Immediate start—once NDA is signed, you will receive repo access and sample bag-of-frames for offline trials. Let me know your relevant projects with RealSense or similar RGB-D systems and your earliest availability.