Research Projects

Robust Nonlinear Adaptive Control of UAV’s using Neural Networks

Mentors: Dr. Bhandari and Dr. Raheja

Cal Poly Pomona (CPP) has been working on a NSF funded project titled “Robust Nonlinear Adaptive Control of Unmanned Aerial Vehicles (UAVs) using Neural Networks.” The overall goal of the project is to develop and experimentally validate neural network based robust nonlinear adaptive controllers for a number of UAVs. As the level of UAV autonomy increases, it is important to design nonlinear controllers that work in the entire flight envelope and take into account the changing flight dynamics. Adaptive controllers help the UAVs adapt to the current situation. We are using different types of neural networks such as multilayer perceptrons, radial basis functions, echo state network, and support vector machine for the dynamic modeling and control of UAVs. The research involves use of flight data for the development and validation of neural network based controllers for the UAVs. The developed controllers are tested in simulation environment, including the software-in-the-loop (SIL) and hardware-in-the-loop (HIL) simulations, and compare the performance of different networks.

Sense & Avoid (S&A) System for UAV’s

Mentors: Dr. Bhandari and Dr. Tang

Due to absence of human pilot onboard, if UAVs encounter some obstacles or other aircraft in their flight path, they cannot take corrective measures to avoid collision. As a result, the crash rate of UAVs has increased significantly. Collision avoidance or S&A capability is one of the preconditions for the integration of UAV’s into the National Airspace System. CPP is currently working on several projects for developing and maturing the technology for UAV S&A capabilities. The research involves investigating the use of various path planning and collision avoidance methods such as Potential Functions, Rapidly-Exploring Random Trees, and combination of these. Different type of sensors such as ADS-B transponders, laser scanners, cameras are used for obstacle detection. The developed algorithms are tested in simulation prior to flight testing.

Coordination and Control of Multiple UAVs

Mentors: Dr. Tang and Dr. Bhandari

CPP has been working on a coordination and control of multiple UAVs. The project uses vision systems onboard the UAVs and other geo-location techniques to determine the target location. The project uses two airplane UAVs and one ground vehicle. The second UAV is used for simulating a rescue mission such as dropping a water bottle to the target. The team of unmanned vehicles can handle more complex task with increased robustness and efficiency through redundancy and task distribution without posing any risk to human life. Our research involves developing operating system and hardware agnostic common control system (CCS) with a specialized, scalable communication architecture that are able to operate multiple unmanned aerial and ground vehicles, and enabling interoperability between vehicles without putting a heavy load on the operator. We will also look into using artificial intelligence and cognitive architectures for the CCS.

Navigation in GPS-Denied Environments

Mentors: Dr. Nakhjiri,  Dr. Tang, and Dr. Bhandari

Autonomous navigation on Earth mostly relies on the Global Positioning System (GPS). However, GPS is susceptible to jamming. Also, GPS signal may be temporarily or permanently unavailable at certain locations. Ability of autonomously navigate in the GPS-denied environments is very important for autonomous operation of UAVs. Computer Vision-Based Navigation (VBN) has potential for UAVs to find the location and direction of the vehicle where GPS coverage is poor or absent. Another technique is to investigate the application of Simultaneous Localization and Mapping (SLAM) techniques for navigation in the GPS-denied environments.  

Indoor Search and Rescue using Unmanned Aerial Vehicles

Mentors: Dr. Bhandari and Dr. Tang

UAVs can be used for search and rescue missions during natural disaster such as earthquake, flood, and fire. Deploying manned aircraft in such situations can pose significant threat to the pilot as well as the aircraft. Moreover, there is no possibility of using manned aircraft when the search and rescue has to be conducted in the indoor environments. It is more challenging to assist victims trapped inside buildings. It requires a quick decision so that as many victims as possible are rescued without risking more lives including those of the rescue teams’. Rotary wing UAVs such as multicopters have potential to be used for search and rescue missions in the indoor environments, which pose unique challenges such as obstacles on the UAV path, the size of the vehicles that can be used, the amount of the payload that the small UAVs can carry, lack of GPS signal and map of the environment, etc.  Multiple small UAVs that coordinate with each other can be used for such scenarios. Small sizes provide the agility and ease of navigation in confined areas. Since these small multicopters are limited in how much payload they can carry, use of two or more vehicles helps in payload distribution. For example, one multicopter can map the interior of the structure and find potential survivors. The second UAV can then be used to deliver a rescue package, which could potentially consist of water, medical supplies, etc.

Unmanned Aerial Vehicles for Precision Agriculture

Mentors: Dr. Bhandari and Dr. Raheja

Remote sensing plays key role in the precision agriculture, which aims to optimize the amount of water, fertilizers, and pesticides. This helps in site-specific management of crops and developing treatment plans.  The images or maps created using remote sensing help understand the environmental effects such as that of weather, water, insects, diseases as well as that of fertilizers and pesticides. Near infrared (NIR) images have been found to be of significant importance in determining the crop performances and stresses. Conventional methods of remote sensing for agriculture applications include satellites and manned aircraft. UAVs can be used cost-effectively for precision agriculture. Cal Pol Pomona has been working on using multispectral and hyperspectral sensors as well as high resolution RGB camera for detection of plant stresses, and treatment. The latter is used for developing machine learning classifiers. Our research focus is on developing a novel technique that will combine UAVs, multispectral images, and machine learning for precision agriculture.

Aircraft System Identification

Mentor: Dr. Bhandari

Aircraft system identification involves using flight data for the identifcation of flight dynamics model parameters. Aircraft (multicopter, airplane, or helicopter) is flown for frequency sweep, doublet, or pulse input in controls. The collected flight data is post processed to remove any noise. The processed data is  then used in the identification of transfer function and/or state-space models for the aircraft using various techniques including frequency response techniques. The identified model response is compared with the flight data of different input shapes for verification. The REU participants will learn methods and types of data collection, flight testing, data processing, and model identification and verification.