1. We introduce RadCloud, a novel framework for unmanned aerial vehicles (UAVs) and unmanned ground vehicles (UGVs) that integrates real-time radar signal processing with a deep learning model to directly convert lower-resolution radar data into higher-resolution lidar-like 2D point clouds in real-time.

2. We optimize our framework specifically for the low-cost processors with limited processing power commonly used by UGVs and UAVs by using a radar configuration with 4x lower resolution and a deep learning model with 2.25x fewer parameters compared to previous works.

3. We utilize a novel chirp-based approach that is more resilient to rapid movements like rapid spins and turns commonly experienced on edge devices such as robots exploring unknown environments.

4. We demonstrate the real-world practicality and feasibility of our framework using commercially available drones and ground vehicles. 

David Hunt - Duke University

Shaocheng Lou- Duke University

Amir Khazraei - Duke University

Xiao Zhang - Duke University

Spencer Hallyburton - Duke University

TIngjun Chen - Duke University

Miroslav Pajic - Duke University

Our paper can be accessed at this link:RadCloud Paper.

Our dataset and code can be accessed at our github repository or by checking out the Code and Datasets  page.