Valeo Schalter und Sensoren GmbH
Kronach, Germany
-
Master's Thesis · Grade 1.0
Extended T-FFTRadNet with a Doppler-aware velocity head, predicting occupancy, object detection and continuous radial velocity together on one dense polar bird’s-eye-view grid. Built DirectFlowMatch to generate the dense velocity ground truth this required. Occupancy TPR improved from 64.5% to 68.0% at a fixed 0.75 confidence threshold. The base architecture and radar dataset were Valeo’s; the velocity head, scene-flow algorithm and evaluation framework were mine.
-
Working Student
20 hours per week alongside my master’s. Investigated whether LiDAR point clouds could be preconditioned for better lossless compression: implemented and benchmarked several strategies against an unpreconditioned baseline, and validated the findings with the author of the codec used. Documented the outcome and preserved the implementation in the codebase.
-
Internship
Full-time, on the “Droid” ADAS-equipped prototype vehicle. Maintained and extended the ROS-based stack on an NVIDIA Jetson under embedded Linux, implemented 4G/5G teleoperation with low-latency UDP transport, and built the four-camera GStreamer pipeline. Most of the work was diagnosing faults across the hardware–software boundary. Demonstrated live to Valeo’s division CTO, regional government representatives and press.