Trajectory Estimation and Motion Planning in Unmanned Aerial Vehicle

International Journal of Advanced Computational Engineering and Networking, ISSN: 2320-2106, Volume-1, Issue-9, Nov-2013 TRAJECTORY ESTIMATION AND MOTION PLANNING IN UNMANNED AERIAL VEHICLE 1ABHISHEK CHAVAN, 2GUNAWANT MALI, 3MANASI DANVE, 4PANKAJ SONAR 1,2,3,4,5Department of Computer Engineering, SITS, Narhe, Pune, India Email: [email protected] Abstract- There is always inquisitiveness in people about flying objects or aircrafts. It is a fascinating thing for an individual to build such a model that can truly fly in air and can be controlled. “Quadrotor” is one of the appealing and interesting designs of a flying object. As per the studies advancements in manual controlled Quadrotor have reached to ultimate extent. Autonomous Quadrotor still has much scope of enhancement. Autonomous Quadrotor needs Trajectory to be defined that is the path to be followed by the Quadrotor. Reinforcement learning strategies can strengthen the power of Quadrotor by progressively improving the trajectory estimation and motion planning. This requires processors with high computational efficiency to process camera captured data and redefine path and motions at every point on trajectory. In this paper we present methodology to optimize processing and reduce computational overheads implementing reinforcement leaning by which continuous monitoring of Quadrotor can be avoided. This will help to explore the potentials of a Quadrotor and can have various real time applications. Keywords: Quadrotor, Trajectory Estimation, Reinforcement Learning I. INTRODUCTION which are enhancing not only the technology but also creating lots of interest in today’s world. At initial Every person has curiosity about the air crafts and its stage, unmanned aerial vehicle and related designs. The research and development in the field of technology were developed to use for military air crafts commenced in 18th century. Scientists purposes. But now, these abilities were used for experimented with various air craft designs to make commercial and governmental service. This them more stable and steady. Etienne Oehmichen was development started increasing slowly and all over the first scientist who experimented with rotorcraft the world. Many countries are focussing on more and designs in the 1920s. He designed and constructed more development in the field of unmanned aerial six multicopter models and tested them. The second vehicles so that they can be used in many different design among these models had four rotors and eight applications. The international as well as the Indian propellers and a single engine was used to drive them. developments are mentioned in the following manner. A steel-tube frame was used along with two-bladed rotors at the ends of the four arms. The angle of these International Developments blades could be varied by warping. Five propellers Recent international quadrotors or quadropters which spinning in the horizontal plane stabilized the are being manufactured and used in aerospace machine laterally. Additional propeller was mounted industry are listed below: at the nose for steering. The remaining pair of AeroQuad is an open-source hardware and propellers was for forward thrust. These models software project which utilizes Arduino boards exhibited considerable stability along with better and freely provides hardware designs and control and took thousand of test flights during software for the DIY construction of middle of 1920’s. The development took pace and by Quadcopters. 1923 it was able to remain in air for several minutes, OpenPilot is a model aircraft open-source and on April 14, 1924 it made the first FAI distance software project. record for helicopters of 360m.Later, it completed the first 1 kilometer closed-circuit flight by a rotorcraft. Developments in India We, as our contribution to this research, aim to give India is also planning to buy UAVs two troops (eight way for a new technique & to add more accurate drones each) of IAI Heron from Israel. Under the Rs artificial intelligence for improving reliability, 1,200crore contracts with Israel Aerospace flexibility, usability and learnability. Industries (IAI), the Army will begin inducting these new Heron drones from January 2014. Also, India is II. LITERATURE SURVEY developing UAVs that are capable of flying on solar power. Developed by DRDO (Defense Research and The survey says that the researches related to aerial Development Organization, New Delhi and vehicle are plenty. Not only manual vehicles but Hindustan, Bangalore).Some of its examples are: autonomous vehicles are on d verge of development DRDO NISHANT ( The Nishant UAV is primarily tasked with intelligence gathering over enemy territory and also for reconnaissance, Trajectory Estimation And Motion Planning In Unmanned Aerial Vehicle 49 International Journal of Advanced Computational Engineering and Networking, ISSN: 2320-2106, Volume-1, Issue-9, Nov-2013 training, surveillance, target designation, artillery Inferences from Literature Survey fire correction, damage assessment, ELINT and This reviewed paper’s presents a comparison between SIGINT) two control approaches for a quadrotor UAV. First, a DRDO NETRA ( The NETRA is an Indian, nonlinear controller based on trajectory estimation light weight , autonomous UAV for has been proposed. Second, a control algorithm has surveillance and reconnaissance operation ) been learned using reinforcement learning and fitted PAWAN UAV’S ( It’s an UAV developed value iteration using the nonlinear dynamic model of for Indian Armed Force ) the quadrotor. Both control engineering approaches result in a satisfying control result. One of the III. LITERATURE REVIEWED advantages of a learning algorithm is the fact that no prior mathematical knowledge of the model is Our system named TEMP is an autonomous required to design a controller. This reflect that, quadrotor which estimates trajectory and plans model of the quadrotor could be approximated using motion on it. We will be using a new concept called different techniques by just relating the input and Reinforcement Learning which will enable our output data through a non parametric approach. This system TEMP to learn itself the path and trace it. For method can be extended to other control approaches this we have reviewed some IEEE papers or used in different types of unmanned aerial vehicles understanding and using them for TEMP. Some of in several applications. Moreover, the proposed the titles of these papers are listed as below: framework has shown to be a good setup for “quadrocoptor trajectory generation and researchers to investigate the application of control” Reinforcement Learning to control algorithm design An algorithm is presented that allows the calculation in particular to UAVs applications. of flight trajectories for quadrocopters. Trajectory feasibility constraints regarding the vehicle dynamics IV. METHEDOLOGIES and input constraints are derived. They are then used in the planning algorithm to guarantee the feasibility Here we would include the methods, the of generated trajectories. The translational degrees of mathematical equations and the mathematical model freedom of the quadrotor are decoupled, and time- of our system TEMP, how these methodologies will optimal trajectories are found for each degree of prove to be possible for our system and how this freedom separately. The trajectory generation is fast could be implemented. enough to be performed online. Control inputs are calculated from the generated trajectory, and used to Forces and Torques on Quadrotor achieve closed-loop control similar to model Forces acting on Quadrotor include lift generated by predictive control. Experimental results showed a each motor and the gravitational pull acting on total good performance, with unmodeled aerodynamic lift generated. Moments can be achieved by effects causing trajectory deviations when generating Yaw, Roll and Pitch movements. Yawing decelerating from high speeds. Development potential Torque is the result of four individual torques for the future is highlighted, focusing on improving generated by spinning motors. Rolling Torque can be the performance and correcting for aerodynamic produced by increasing left motor’s thrust and effects. decreasing right motor’s thrust and vice versa. Pitching Torque can be generated by increasing front “An Experimental Validation of Reinforcement motor’s thrust and decreasing back motor’s thrust and Learning Applied to the Position Control of UAVs”. vice versa. An algorithm involving reinforcement learning technique is being implemented in this paper. The Reinforcement Learning Methodology algorithm states that by iteratively calling the same Reinforcement learning is the branch of computer task, one can enhance the performance of the science that enables machine leaning. Self leaning Quadrocopters. The algorithm also focuses on how strategies are being used to make machines smart and effective would it be to use the technique in different intelligent. Using appropriate algorithms machines sense. Different inputs are taken and controlled can recognize the environment and take precise outputs are obtained in it. The positions of a decisions without any manual inference by humans. Quadrocopters are well judged and defined and Machines can be made capable to analyze and adapt accordingly steps are taken to acquire it. Different to the environment to perform the expected task in a angles required different measures taken to gain the managed way. The Reinforcement learning system positions

Load more