Fraunhofer Industrial AI, 3D Vision & Robotics

Computer Vision and Robotic Engineer at Fraunhofer Innovation Platform

Role and Mission Overview

At the Fraunhofer Innovation Platform, I work on autonomous mobile robots, robotic manipulator arms, production lines, and advanced manufacturing systems. My focus is bringing AI solutions to increase line flexibility and resolve technical bottlenecks for fast adaptability, helping bring Industry 5.0 concepts into real-world factory floors.

Oct. 2025 to Present Enschede, Netherlands

Real-Time Battery Anomaly Detection & Multi-Modal Inspection

The Challenge

In high-speed battery production, minute surface defects, terminal misalignments, or internal electrochemical irregularities can cause cell failure. Manual inspection is slow and prone to fatigue, making it difficult to keep up with active conveyor lines.

The Solution and My Role

The complete inspection pipeline integrates high-speed industrial cameras for surface defect detection, electrochemical testing for internal resistance and voltage validation, and a Yaskawa robotic arm to sort and move cells. I built the AI monitoring system for real-time anomaly detection and programmed the Yaskawa robot arm to handle, inspect, and route individual battery cells across the line.

95%
Defect Recall Rate
Catches visual and structural flaws reliably
50%
Cycle Time Saved
Faster in-line multi-modal inspection throughput
Multi-Modal
Vision + Electrochemistry
Synchronized optical and internal resistance testing

Flexible Industrial Robotic Assembly and Manufacturing

The Challenge

Modern modular manufacturing and circular repurposing require robots to perform diverse operations on changing product models. Hardcoded routines struggle when shifting between different pack shapes, cell dimensions, and assembly stages.

The Solution and My Role

I programmed Universal Robots (UR) and Yaskawa robotic arms to execute a wide variety of operations across the manufacturing cycle, including pick-and-place, precision screwdriving, welding, and in-line part inspection. The system dynamically adapts based on the current required task and end-effector tool without requiring line re-tooling.

Pack Pick, Place & Automated Screwdriving

Continuous Multi-Robot Manufacturing Flow

Modular
Different Tools & Tasks
Pick-and-place, welding, screwing, and inspection
2 Arm Types
UR and Yaskawa
Integrated multi-robot production workflow
Adaptable
Fast Reconfiguration
Adapts quickly to active product variations

Sub-Millimeter 3D Manipulation with Zivid and Pick-it

The Challenge

Harsh industrial setups present severe challenges such as visual occlusions, fluctuating ambient light, and unpredictable part placements. In flexible manufacturing lines, raw incoming materials are not always locked into pre-defined fixtures and can shift anywhere from millimeters to several centimeters, which would normally break rigid robot cycles. In addition, reflective metal parts create specular glare that defeats standard 2D cameras.

The Solution and My Role

I integrated Zivid 3D structured-light cameras and Pick-it 3D vision processors into the robotic control loop. The pipeline filters dense spatial point clouds and calculates real-time 6D grasping poses, allowing the robot to accurately locate and manipulate parts despite occlusions, specular reflections, and position variations.

< 0.8 mm
Grasping Precision
Verified sub-millimeter accuracy under harsh reflections
Dynamic 6D
Pose Estimation
Compensates for multi-centimeter part shifts in bins
Occlusion Safe
Vision Engine
Zivid + Pick-it point cloud filtering under industrial clutter

Instruction-Guided VLA Controllers (Ongoing Research)

The Challenge

Making industrial robots easier to use on the factory floor means allowing operators to give simple typed or spoken instructions without writing robot motion code.

The Solution and My Role

I am conducting ongoing research on Vision-Language-Action (VLA) controllers. We are collecting multimodal demonstration datasets and building policy models that translate operator text instructions and camera views into robot trajectories.

Multimodal
Vision, Language, Action
Direct text command grounding into physical trajectories
Zero-Code
Operator Interface
Enables workers to direct assembly flows naturally
Active
Ongoing Development
Dataset curation and policy training pipeline

Technologies and Related Projects

Zivid 3D Camera Pick-it 3D Universal Robots Yaskawa Motoman PyTorch ROS2 Vision-Language-Action PCL / OpenCV