01
Master's Thesis · Radar Perception
End-to-End AI Radar Signal Processing
Extended T-FFTRadNet with a dense radial-velocity head for joint detection, occupancy, and per-cell velocity prediction from range–Doppler radar.
Aditya Ramchandra Pataki
Robotics software & Perception Engineer
I work across radar, LiDAR, ROS/ROS2, and embedded robotics, combining learning-based perception with hands-on systems integration. I’m especially interested in teleoperation, trajectory and motion planning, connected automated vehicles, and V2X systems.
Radar · LiDAR · ROS/ROS2 · CAVs · V2X · Autonomous systems
With appreciation for the engineering and research teams at Valeo whose collaboration supported and shaped this work.
LocationErlangen, Germany · Open to relocation across the EU
Work authorizationGerman work permit, no sponsorship required
AvailabilityAvailable immediately
01 / Selected work
Engineering work, explained through decisions and evidence.
Four case studies across radar perception, LiDAR scene flow, robotics systems integration, and point-cloud evaluation—each focused on the problem, my contribution, and the outcome.
02
LiDAR Scene Flow
DirectFlowMatch
A deterministic, geometry-aware scene-flow method that estimates per-point motion from consecutive LiDAR scans without segmentation or clustering.
03
Human-in-the-loop robotics
Teleoperation Systems Integration
Implemented 4G/5G teleoperation on an ADAS-enabled miniature vehicle using NVIDIA Jetson, ROS control, four-camera GStreamer feedback, and end-to-end reliability debugging.
04
LiDAR Data Investigation
Point-Cloud Compression
Researched several preconditioning techniques for 3D point clouds to improve their compressibility, from image-based representations to structure-aware ZFP experiments.
Loading the Bunny comparison…
02 / Work visualization
A vehicle-rich nuScenes trace, from two scans to complete motion.
This high-motion public-data trace makes moving traffic and flow direction inspectable: compare the input pair, see the initial matching, then play the complete point-level match.
nuScenes · High-motion traffic trace
From two scans to one complete point-level match.
Inspect the input pair, initial matching, and complete match.
Active view 00 · Input pair Two consecutive, voxel-downsampled LiDAR scans.
Preparing the interactive DFM trace…
03 / Technical stack
Tools for sensing, learning, and deploying robotic systems.
A focused stack grounded in the work shown across the case studies.
01
Robotics
- ROS
- ROS 2
- GStreamer
02
Perception
- Radar
- LiDAR
- Computer vision
03
Learning
- PyTorch
- Transformers
- BEV prediction
04
Systems
- NVIDIA Jetson
- Linux
- C++
- Python
05
Engineering
- CMake
- Git
- Technical documentation
06
Connected autonomy
- Trajectory planning
- Motion planning
- CAVs
- V2X
04 / Selected outcomes
Four signals of practical engineering depth.
Results from research, systems integration, and technical evaluation.
Presented to management, local-government representatives, and press.
Fixed 0.75 confidence threshold on highway evaluation.