Available from 1 January 2027

Batchaya Yacynte —
Perception engineer.

Building the vision systems that let machines navigate when GPS fails. M.Eng. AI Engineering & Autonomous Systems · TH Ingolstadt · Airbus Helicopters consortium.

See my work ↓ IEEE Publication ↗ Download CV ↓
Batchaya Yacynte
4+
Years building
perception systems
0.15°
Rotation accuracy
Project ENGEL (simulation)
1st
Author IEEE publication
REM 2024
3
Real hardware platforms
validated outdoors

Affiliated with

TH Ingolstadt - M.Eng. Airbus Helicopters Consortium Fraunhofer IVI AIMotion Bavaria IEEE REM 2024 - First Author

About

Engineer, researcher,
builder.

I'm a Cameroonian robotics engineer based in Ingolstadt, Germany. My work lives at the boundary where computer vision, probabilistic estimation, and control systems meet, building perception pipelines that have to perform on real hardware, outdoors, under pressure.

My master's thesis, developed within an Airbus Helicopters consortium, returns a UAV camera to the viewpoint of a reference image using only an onboard monocular camera. In 234 Unreal Engine 5 simulation trials across clear weather, rain, snow and fog it reaches 0.15° rotation accuracy and about 2 px image alignment in clear weather; first tests on the real drone cover the gimbal (rotation) loop. Before that, at Fraunhofer IVI, I designed an event-based optical flow system for Time-to-Collision estimation that outperformed three state-of-the-art methods on the MVSEC benchmark, running single-threaded on NVIDIA Jetson.

I care about systems that work outside the lab. My work is validated on KITTI, EuRoC, custom DJI Tello flights with OptiTrack ground truth, and live outdoor field tests, not just in simulation.

Selected work

Projects that ship.

Real-world perception systems, embedded deployments, and autonomous platforms, from Airbus-consortium UAVs to Fraunhofer agricultural robotics.

01
Master's Thesis Airbus Helicopters Consortium · AIMotion Bavaria 2025 – 2026
Project ENGEL: Closed-Loop Vision-Based UAV Repositioning
Closed-loop monocular repositioning: after GNSS brings a UAV within a few metres of a recorded viewpoint, the system drives the vehicle and gimbal back to the exact view of the stored reference image. SuperPoint + LightGlue matches with sub-pixel refinement and uncertainty feed a PNEC rotation estimate; translation is measured as a signed image error, and saturating control laws close the loop at ~15 Hz (C++, ONNX/CUDA). Built the Unreal Engine 5 test environment from Google 3D Tiles with dynamic sky, rain, snow and fog. Results below are from simulation; first real-drone tests cover the gimbal loop only.
0.15° rotation error (sim, clear) ≈2 px view alignment (sim) 234 sim trials · 6 weather conditions ~31 s per run (sim) Real drone: gimbal loop tested
Repositioning code ↗ Controls ↗ UE5 weather ↗
Project ENGEL repositioning result in simulation Rotation error over time during repositioning Translation error over time during repositioning
ENGEL · simulation
↗
02
Research · KoSiNuS Fraunhofer IVI Apr 2025 – Mar 2026
Event-Based Optical Flow for Real-Time TTC Estimation
Probabilistic event-based optical flow pipeline for Time-to-Collision estimation in safety-critical agricultural autonomy. Gaussian residual model with motion-adaptive variance scaling and affine RANSAC outlier rejection, integrated into ROS2 with an IDS uEye DVS sensor on NVIDIA Jetson embedded hardware.
2.33 px AEPE Outperforms UnFlow, EV-FlowNet, Shiba et al. 3.5 Meps single-threaded on Jetson Orin
Event-based optical flow compared with ground truth
KoSiNuS · Event Flow
–
03
Personal Project Custom UAV Build
Autonomous Surface Area Mapping Drone
Designed and hand-built a custom ~25×25 cm UAV (Betaflight F405 V3 + Radxa A7A) and developed a full stereo vision pipeline for autonomous land surface area measurement. Targets agricultural and surveying use cases as a low-cost alternative to total stations. Full pipeline from stereo depth estimation to contour detection and geometric area computation, generating topographic and plat map outputs.
KITTI validated Topographic + plat map outputs Custom hardware build
Mapping code ↗ Flight control ↗
Beetle, the custom-built mapping drone Beetle drone hardware close-up Topographic map produced by the stereo pipeline Plat map with computed surface area
Beetle · Maps
–
04
Course Project Grade 1.0 2025 – 2026
Deep Visual-Inertial Odometry Pipeline
End-to-end deep learning VIO system for GPS-free 6-DoF localization, fusing monocular camera and IMU measurements with CNNs. Evaluated on EuRoC MAV for indoor flight and KITTI for outdoor driving scenarios.
0.038 m ATE — EuRoC MAV 1.42 m ATE — KITTI
Code ↗
Deep VIO estimated trajectory on KITTI Deep VIO trajectory compared with ground truth Deep VIO estimated trajectory on KITTI Deep VIO trajectory compared with ground truth
Monocular · VIO · Learning
–
05
Robotics Engineering CAD & Mechatronics Personal / Industrial
Robotic Systems & Mechanical Design
Mechanical and mechatronic engineering projects spanning robotic systems, CAD design, vehicle assembly-line concepts, and agricultural robotics. Designed mechanical assemblies and integrated them with electronics, actuators, sensing, and autonomous control concepts.
CAD design Robotic assemblies Agricultural robotics Mechatronic systems
Vehicle assembly-line concept CAD Wheel assembly CAD model Gearbox CAD model Agricultural seeding robot prototype
Robotics · CAD
↗
06
Bachelor's Thesis IEEE REM 2024 · First Author 2022 – 2024
Monocular Visual-Inertial Odometry for GPS-Free Drone Navigation
Sparse optical flow VIO pipeline for autonomous drone navigation without external positioning. RANSAC-based outlier rejection and camera-IMU calibration. Validated across KITTI benchmark, DJI Tello flights with OptiTrack ground truth, and Unreal Engine simulation.
0.94% translation RMSE/m — KITTI 5.32% — DJI Tello real flight 0.015 deg/m rotation error
Code ↗ Tello + OptiTrack ↗
Simulation result 3D Translation result Tello flight Optical flow
Monocular · VIO · Geometry
↗

Publication

Peer-reviewed research.

IEEE REM 2024 · First Author · DOI: 10.1109/REM63063.2024.10735689
Getting Started With a Simple Visual-Inertial Odometry
Batchaya, Y. D. et al. — Presented at IEEE REM 2024 · Published in IEEE Xplore
IEEE Xplore ↗

Contact

Let's build
something real.

Looking for robotics and computer vision roles in Germany starting 1 January 2027. Eligible for an EU Blue Card or the post-study job-seeker permit. If you work on autonomous systems, UAVs, or embedded perception — let's talk.

CV Batchaya_Yacynte_CV.pdf ↓ Email batchaya.yacynte@gmail.com → LinkedIn linkedin.com/in/batchaya-yacynte → GitHub github.com/Yacynte → Publication IEEE REM 2024 · DOI 10.1109/REM63063.2024.10735689 →