Robotics and Autonomous Systems (RAS) such as robots and Unmanned Aerial Systems (UAS) have been widely deployed in today’s industrial applications, and it has been demonstrated that RAS can replace labor workforce through monitoring (perception) and maintenance (actuation) for many jobs such as human-machine interaction, autonomous driving, advanced manufacturing, cleaning, and so on. For solar Photo-Voltaic (PV) systems, monitoring and maintenance job is labor-intensive. Also, most of PV systems are installed at remote areas or building rooftops where they are not easy to reach. Thus, it is costly – and some time, even dangerous -- to assign labor workforce to perform PV inspection and maintenance because of the high risk and extensive site work. In the particular case of Middle East and North Africa (MENA) region, such as Qatar (Kazem and Chaichan 2019) the unique desert climate condition, and the issues of soiling and extremely hot outdoor temperature make PV maintenance even more challenging than other areas in the world. The strategic goal of this project is to design and develop a novel, hybrid UAS for tackling PV inspection and cleaning tasks. The socio-economic interests include inspection of large PV installations, close-range PV panel inspection using thermal imaging, brush-based drycleaning for dust removal. Thanks to the development of RAS technologies, there have been several successful robot deployments in the field of solar PV system monitoring and maintenance (Iqbal et. al 2019). For PV cleaning, there are several automatic cleaning machine products in the market. The most common PV cleaning machines consist of a fixed structure that must be installed together with the PV arrays (see Fig. 1), which makes them inefficient in terms of cost when the PV array is not installed in a regular layout (Al Shehri et. al 2016). Furthermore, the cleaning machines are not able to autonomously adapt for condition-based maintenance, e.g., weather conditions such as soiling and rain that are different throughout the year. More portable solutions are also available in the market (see Fig. 2). However, these solutions require the cleaning robot to be placed on top of the panels, either using an additional ground robot with a manipulator or a human operator. For PV inspection, currently there are several RAS solutions such as UAS with cameras (Díaz et. al 2020). Using image data collected at a high altitude, the system can assess the condition of PV farms and it can save site workers a lot of workload comparing with routing visual inspections performed by human operators. While traditional methods for PV inspection capture the entire region at several meters above the panels, close-distance thermal imaging by drone infrared cameras can detect PV module defects in a more effective and fast way (Jahn et. al 2018). A hybrid drone that routinely cleans PV panels, autonomously land and move on top of PV panels, and uses a thermal camera to capture close-distance images can not only meet the requirements for frequent and automatic scan, but also has advantages such as closer image-capture distance, lower requirements for onboard cameras, and acquisition of high-resolution images. Motivated by the crucial need for PV monitoring and maintenance under Qatar climate conditions, this project will develop an integrated RAS solution for PV monitoring and cleaning, which can improve PV maintenance efficiency and reduce the labor workforce in PV maintenance tasks. Based on dual thermal imaging based on Electro-Luminescence (EL) and Photo-Luminescence (PL) cameras, onboard of the drone, a more accurate PV inspection tool will be developed and demonstrated for PV reliability monitoring scenarios in Qatar. The drone will also have the capability of performing adaptive cleaning tasks on the PV panels’ surfaces, achieving a high-performance dust removal to fit the climate variations under Qatar weather conditions, and also increase the electricity production due to highefficiency maintenance of PV systems. A number of hybrid flying robots designs have been proposed for a variety of applications (Page and Pounds 2014; Mintchev and Floreano 2018; Nishimura and Yamaguchi 2020; Atay et. al 2021; Sarkis et. al 2022). They continue to draw attention due to the high maneuverability, speed, and versatility, with respect to their single locomotion counterparts. To the best of our knowledge, the design proposed by (Sarkis et al. 2022) is the only hybrid drone specially designed for PV cleaning tasks. Their design showed promising results, yet it is rather simplistic and the ability to roll over the PV cleaning is not tested thoroughly. Furthermore, it does not address PV inspection tasks. This proposal is an exploratory work towards a more efficient hybrid RAS solution specifically designed for Qatar weather conditions. The focus will be on a combined RAS solution that integrates both UAS-based thermal imaging with a portable cleaning robot. An integrated solution would provide the advantages of agile mobility and rapid deployment that UAVs provide with the convenience of portable cleaning robots (Sarkis et. al 2022). We aim at leveraging the recent advances in hybrid drones to propose an innovative solution to PV cleaning and inspection. The project will tackle both modeling, design, control aspects of the hybrid drone. For the design part, we will expand an existing software tool (Xu et. al 2019) aimed at non-conventional drone design. The software tool lets the users try different drone designs, allowing them to choose from a selection of parts for the frame, propellers, etc. The tool also includes a physics-based simulator component, that can be used to test flight controllers to determine whether the drone is able to take off, fly, and land. In this project we will adopt a reinforcement learning (RL) approach to develop the controller for the hybrid drone. A reinforcement learning approach (Sutton and Barto 2018) has been chosen mainly because it is model free and it has been proven to be very effective for hybrid UAVs (Xu et. al 2019). We will adapt the Learning to Fly tool proposed in (Xu et. al 2019) to include wheeled locomotion, which will enable the design, modeling, and simulation of hybrid drones that can fly and move on the ground using wheels. Based upon that, a neural network controller will be trained by means of simulation. As a last step, the trained controller will be ported to hardware. Project impact: The proposed hybrid drone will represent a novel solution to the problem of PV cleaning and inspection in irregular PV installations. Project’s outputs will be innovative and original contributions to the recently developing field of hybrid UAVs. It will also provide valuable insight on the implementation of neural network controllers for these type of hybrid robots. Moreover, up to our knowledge, our system will be the first application of reinforcement learning to hybrid ground/air robots. We expect that the trained controller, once ported to the real drone, will show remarkable efficiency and accuracy. As a result, the entire drone development pipeline will prove to be effective and will open the door to other hybrid robotic designs that will have a clear socio-economic impact, in particular for Qatar’s renewable energy industry.