Academic Projects

Multi-Agent path planning for future Air Traffic Management

Air traffic control is currently operating close to saturation. The amount of traffic is expected to double by 2025. New decision tools to provide Traffic Flow Managers adequate information are needed to handle the upcoming increase in traffic and diminish its environmental impact. This paradigm shift is sustained by a global effort, namely NextGen in the US and SESAR in Europe.

In this project my partner and I explored Multi-agent path planning algorithm for future Air Traffic Management. Especially, we modeled Air Traffic Management constraint for grid-based planning. Then, we implemented A*-based multi agent algorithms, named CA*, HCA* and WHCA*. Moreover, we tried to fuse HCA* and direction maps (DM). Direction Maps is an heuristic that promotes trajectories that go with the general consensus. PDF

A Model Predictive Control Approach for Active SLAM

A central problem in deliberative robotics is mapping an unknown world. For instance, search and rescue robots enter an unknown disaster area and must generate a map as they explore. Simultaneous Localization and Mapping (SLAM) is a successful technique for mapping an unknown environment with noisy sensors. Usually, acquiring sensor data over the whole map requires human intervention to cover unexplored area. This process does not scale as intelligent robots become ubiquitous. Beyond the fact that offline algorithms cannot determine a strategy to completely cover the map, the computed path is clearly not optimal.

Active SLAM is a particular issue of Active Sensing, which address the problem of how to optimally control a sensor for information gathering. The aim is to obtain a policy which trades off exploration (maximizing map coverage) and exploitation (going back to known locations to increase the accuracy of the map). In this project, we proposed an approximate method to solve the intractable Active SLAM problem. We implemented a Model Predictive Controller and compared it to the Greedy approach. PDF

Markov Random Field segmentation for traffic sign detection

Traffic sign detection and recognition have received an increasing interest in the last few years. This is due to the wide range of applications: Highway Maintenance, Sign Inventory, Driver Support Systems, and Intelligent Autonomous Vehicles. Sign detection involves locating road signs within images, and sign recognition involves extracting the sign type. The goal of detection is to extract the signs from the background scene. In this project, my partner and I focused on the segmentation step involved in the first stages of traffic sign detection.

We chose to use Gaussian Markov Random Fields, a model based approach to texture segmentation, with a well-chosen set of features computed from Gabor filters. Moreover we tried both a supervised and unsupervised segmentation. Unsupervised segmentation is attractive for its robustness to change in scene illumination, because is too computational to be apply in production. PDF