ASPIRE: Towards robust perception of radar images

Radar images are often disturbed by noise, clutter, and interference, which sometimes leads to a misinterpretation of those radar images. Better understanding of the impact of noise and interference on radar analysis will lead to more robust interpretation of radar images and more reliable radar-based Advanced Driving Assistance Systems (ADAS) applications
 

Assignment

In this research program we have the following objectives:

1.           To quantify the impact of radar clutter, noise and interference on the interpretation of radar images, design statistical models of those radar jamming and use those models to generate realistic synthetic radar images

2.           To use such synthetic radar image datasets to train ML models for radar image interpretation for specific ODDs, potentially yielding more reliable interpretation

3.           Rigorously assess and benchmark different ML models for radar image interpretation for static and dynamic scenes that can adapt to specific ODDs (e.g. self-parking and autonomous driving perception)

4.           Integrate robust and adaptive ML pipelines into ADAS simulation stack and design new ADAS algorithms that rely on radar and can handle the imperfections in radar data (noise, clutter, interference patterns) for different ODDs (e.g. weather).

5.           Optimize the interface between the ADAS algorithms and human users by taking advantage of the uncertainty in the radar interpretation. Specifically, to explore user adoption and improve the collaboration between the ADAS algorithms and human users for symbiotic driving.

The MSc student shall conduct research along the lines of one of those objectives, co-supervised by PhD students and professors from multiple TU Delft faculties, and engineers from NXP.

Contact

Last modified: 2025-05-19