AI-Enabled Tactical Route Planning --- Student Researcher

To simulate the real-world drone tracking and path planning process, I use air-sim together with Unreal Engine to simulate the car tracking process. Firstly, we built a true-to-real-world environment in Unreal Engine (UE4), as shown in the first picture, which reduced the cost to do the real physical experiments. Then I implement a deep-learning-based depth-map algorithm (Monodepth2) and segmentation algorithm(Detectron2) in the virtual environment. These two methods will generate the 3D occlusion identification and map it into the Unreal Engine virtual environment. Furthermore, we used RRT algorithms for the 2d path planning (Fig 2). We set the starting point (X) and a target point (O). The red line is the optimal path and the green line is all possible paths (Fig 3). After getting the optimal path from above, we change the path into spline in UE4. The planned path in Fig 3 is executed by the virtual agent in Fig 4. In the last step, we designed a spline-based autonomous moving method for agents using a blueprint that also could record real-time position data in TXT format. The recorded position data will be used for the drone to track. This technology can be used for future autonomous driving as drones can assist the car in driving and path planning.

Human-Robot Collaboration System (LMCO project) for Target Tracking --- Team Lead

First of all, I use YOLO for target detection (Fig 1). The target data will be stored in the database first. I predefined multiple positions of virtual human targets as shown in Fig 2. I also put several fixed-camera in the environment which provides views for human operators. When drones find the target, they will track the target (Fig 3 – 4) using a Neural Network based trajectory prediction algorithm. When drones lost targets due to bad weather or obstacles (Fig 5), the drone can receive messages from the human operator (Fig 6) through the fixed-camera windows, which assists the path planning for the drone. Drones will update the path based on the click position. I also designed a fog system in the environment which means we can simulate the drone’s tracking ability in different densities of fog.

Research on Volumetric Contact Theory to Electrical Contact between Random Rough Surfaces --- Team Lead

The core problem of metal contact was to calculate the contact area accurately. Traditionally, researchers tend to use fractal theories, while the applications are limited in some cases. Inspired by holes in sponges, I used the concept of porosity to develop a universal methodology. The big picture was to divide the volume of the metal by the height of the refactored surface. With the reconstructed surface, I wrote a program to iterate the height of the cylinder mentioned above recursively. Consequently, the real contact area can be calculated. Luckily, my supervisor, Professor Wurui Ta, got contact with Dr. BNJ Persson – who raised Persson’s contact theory, which becomes one of the most important theories in contact mechanics – to evaluate the model I developed. Combining with Persson’s method, we also propose a new resistance calculating method which could be used to determine superconducting cables’ resistance quicker and cheaper. The simulation results show better accuracy and speed than any existing method. This research was published in Tribology International