Autonomous Drone for Search & Rescue

12/2015 – 12/2016

This project is centered around a theoretical search-and-rescue mission. The task is building an application that enables a drone to

autonomously take off from a moving vehicle,
collect data on the whereabouts of survivors and transmit the information from the disaster site back to the vehicle, then
land autonomously on the same moving vehicle.

Target Detection module: developed a real-time Vision-based module (including Object detection and 3D localization) on Drone’s autonomous navigation system. Real-time 3D poses of objects are estimated from 2D images using the real-time camera gimbal data based on Homography and 3D Geometry.
Go-to-goal module: developed an autonomous tracking-and-landing module for autonomous landing on a still/moving vehicle.
Developed a framework managing multimodal sensors, which makes decisions from real-time feedbacks collected from cameras, drone states, and others sensors, duing Drone’s navigation
Information display on APPs.
Designed and Setup the testing environments (collaborating with the team) to evaluate the Drone’s mission.
Conducted 100+ outdoor tests using Matrice 100 and real trucks (collaborating with the team), tuning the hyperparameters to guarantee drone’s landing safety.

Achieved ≤10cm autonomous landing accuracy.
Won the 4th Place in World Top10 in the International Contest of “2016 DJI Drone Developer Challenge” taking place in New York, USA. There were a total of 130+ international participants. The contest was hosted by United Nations Development Programme, DJI, and Ford. A video is available here.
Later, we extend the demo to another project that involves communication & interaction between Drone and Ground Vechile, allowing a collaboration between aerial autonomous robots and ground autonomous robots.

Instructor: Nanning Zheng (Institute of Artificial Intelligence and Robotics @ Xi’an Jiaotong University)
Affiliation: Xi’an Jiaotong University
Teamates: Yuchan Liu, Shitao Chen