Ashwin Joseph — Radar Perception, Sensor Fusion & Robotics Research Engineer

About

Working at the boundary of perception and simulation

I'm Ashwin Joseph — a Research Associate at Fraunhofer FHR in Bonn, where I work on radar perception, sensor fusion, and electromagnetic digital twin simulation — building ray-traced radar scenes with NVIDIA Sionna RT to model how automotive and robotic radar systems perceive complex environments before real hardware is involved.

My path here runs through two M.Sc. degrees — Robotic Systems Engineering at RWTH Aachen and Computer Science / AI at CUSAT, India — and industry stints at Sony India and Ignitarium (Neurealm), where I shipped AI camera modules on Qualcomm hardware and built real-time defect detection systems. My master's thesis at DLR Neustrelitz focused on Visual-LiDAR SLAM for autonomous waterway navigation.

I'm drawn to problems where simulation, signal processing, and robot learning intersect. I'm currently exploring World Models, VLA-based manipulation and autonomous navigation research in my free time. If you're working on radar simulation, multi-sensor perception, or robot learning, I'm always happy to connect.

Vehicle Pedestrian Road Building Vegetation

Semantic point cloud // automotive scene

Focus areas

What I spend my cycles on

Radar Perception & Sensor Fusion

Multi-sensor fusion pipelines combining automotive radar, LiDAR, and camera data for semantic 3D mapping and scene understanding.

EM Digital Twin Simulation

Ray-traced radar channel simulation with NVIDIA Sionna RT — building indoor scenes like warehouse in Isaac Sim and Blender, for physically accurate propagation modeling.

SLAM & Autonomous Navigation

Visual-LiDAR SLAM and occupancy mapping for autonomous vehicles.

Robot Learning

Building toward VLA and manipulation research — working through the ETH Zurich Robot Learning curriculum alongside applied radar work.

Experience

Track record

01

Research Associate Current

Fraunhofer FHR — Bonn, Germany

Multi-dimensional FFT-based radar processing — range-Doppler processing, spatial FFT for angle estimation, windowing, adaptive CFAR detection, and coordinate transforms to generate 3D radar point clouds. Contributing to electromagnetic digital twin development for indoor environments using NVIDIA Sionna RT, Cosmos, and Isaac Sim, and working on radar-based perception with the nuScenes dataset for automotive use cases.
02

Master's Thesis and Internship — Robotics

German Aerospace Center (DLR), Neustrelitz

Semantic 3D mapping for autonomous inland waterway vehicles using Visual-LiDAR SLAM and GNSS/stereo camera/LiDAR fusion. Work now forming the basis of a paper submitted to IEEE OJ-ITS.
03

Consultant — AI (on deputation)

Sony India Software Center

Deployed AI-based camera modules for Sony Android devices (Xperia); built benchmarking pipelines on Qualcomm Snapdragon hardware for image super-resolution, object detection, and white balance correction.
04

Senior Engineer — AI

Ignitarium Technology Solutions (Neurealm)

Led predictive maintenance, defect detection, and real-time video analytics for manufacturing and railway safety applications, including a railway joint-bar fault detection system using IR depth cameras. Mentored junior engineers and handled client deployments and executive demos.
05

ML Contractor (on deputation)

OpenFive Technologies (SiFive USA)

Ported custom neural networks onto FPGA, optimized models with CEVA LiteML, and ran performance benchmarking for real-world edge deployment.

Selected work

What I've built

Electromagnetic Digital Twin Pipeline for Radar Simulation

An end-to-end pipeline for physically accurate radar simulation: scenes constructed in Isaac Sim with SimReady assets, then ray-traced with NVIDIA Sionna RT — from full warehouse environments down to single-vehicle studies. Includes a custom Plotly-based 3D visualizer for path-level analysis by amplitude, interaction type, and angle of departure.

$Material-aware ray-traced radar simulation of a car in an indoor scene, showing specular, diffuse, and refraction propagation paths between transmitter and receiver with per-material scene classes$

Material-aware ray tracing of a vehicle in an indoor scene at 5 GHz — 299 propagation paths between transmitter and receiver, split into specular (yellow), diffuse (violet), and refraction (teal) interactions, with per-material modeling of body panels, glass, tires, and interior.

Sionna RTIsaac SimBlenderRay TracingMaterial ModelingRadar Simulation

Perception in action

3D Semantic-Enriched Mapping

Sensor fusion building up a semantically labeled 3D occupancy grid around the ego platform — the same principle behind the nuScenes occupancy prediction work and the inland-waterway semantic mapping thesis, applied here to a driving scene.

Predictive Maintenance for Railways

Deep learning pipeline using IR depth cameras to detect joint-bar faults along railway tracks, for GREX (USA).

Footwear Defect Detection

Vision inspection system for raw-material quality control, for VKC (India) — cut test time from minutes to 20 seconds.

Watch demo →

Voice Denoising & Keyword Spotting

Denoising and keyword-detection techniques for voice-based deep learning, with synthetic training data, for Renesas (Japan).

Watch demo →

Attire Detection & Validation

Human pose estimation and custom object detectors for attire compliance checks, for Ericsson (Sweden).

Watch demo →

UR10e Cobot Trajectory Planning

Welding-tip trajectory planning for a UR10e collaborative robot using MoveIt.

Watch demo →

Humanoid Manipulation Hackathon

Dual-cobot humanoid setup on a Franka Emika manipulator — VR teleoperation via Meta Quest 3 and imitation-learning policy training/deployment.

Watch demo →

Generative AI & Synthetic Data

Autoencoder, diffusion, and GAN experiments, plus Blender-based synthetic data generation with domain randomization.

AshwinJoseph

Working at the boundary of perception and simulation

What I spend my cycles on

Radar Perception & Sensor Fusion

EM Digital Twin Simulation

SLAM & Autonomous Navigation

Robot Learning

Track record

What I've built

Electromagnetic Digital Twin Pipeline for Radar Simulation

3D Semantic-Enriched Mapping

Predictive Maintenance for Railways

Footwear Defect Detection

Voice Denoising & Keyword Spotting

Attire Detection & Validation

UR10e Cobot Trajectory Planning

Humanoid Manipulation Hackathon

Generative AI & Synthetic Data

Writing & research

Out in the world

Demoing AI attire detection to the Minister of State for Electronics & IT

Senior mentor, "Ask the Seniors" week

Toolkit

Languages & Frameworks

Robotics & Simulation

Deployment & Tools

Specialties

Foundations

RWTH Aachen University

IIITM-K, Trivandrum

Mahatma Gandhi University