Navigation ensures drone can travel from one waypoint to another. The nav- igation task requires drone to first localize itself to known map and estimate its position and velocity. Then it must apply thrust to fly towards its desti- nation. The obstacleavoidance library developed above can be extended for navigation as well. Additional sensors which can assist in navigation can be added to sensor modules. An example of such sensors is visual odometry or optical flow sensor which was discussed in chapter 2.1.3. Sensor fusion algo- rithm can extend or override the algorithm provided by obstacleavoidance library for supporting navigation. The extended algorithm needs to plan the navigation path in addition to keeping track of the obstacles. Mavlink com- munication module must implement additional mavlink messages for sending a command to flight controller and navigate based on the relative position or relative attitude. Some of such commands which are used for navigation are SET POSITION TARGET LOCAL NED and SET ATTITUDE TARGET. The use of optical flow can also help in mitigating problem of flying in an area with bad GPS signals such as forests, cities with tall buildings and tun- nels. Artificialintelligence can also provide assistance for auto navigation of the drones in such places.
The field of creation of intelligent machines that work like humans and respond quickly, in computer science is known as Artificialintelligence. The core part of AI research is Knowledge engineering. Machines can react and act like humans only when they have abundant information related to the world. To implement knowledge engineering, Artificialintelligence should have access to objects, categories, properties, and relations. To initiate common sense, reasoning and problem-solving power in machines, it is a difficult and tedious task. Machine learning is another one of the core parts of AI. Learning without any kind of supervision requires an ability to identify patterns in streams of inputs, whereas learning with adequate supervision involves classification and numerical regressions. Classification determines the category an object belongs to and regression deals with obtaining a set of numerical input or output examples, thereby discovering functions enabling the generation of suitable outputs from respective inputs. Mathematical analysis of machine learning algorithms and their performance is a well-defined branch of theoretical computer science often referred to as computational learning theory.
Visually impaired individuals find it more challenging to move about independently because of their compromised vision. Moreover, a blind person’s capacity to navigate in a given setting, along with their ability to organize their daily activities are vital to their health and wellbeing. Organizing any commonplace activity can be especially difficult for a blind man/woman if he/she has not learned to distinguish between different items like drug containers and packaged goods, just by feeling with the hands. The more saddening fact is that there are tens of millions of visually impaired persons worldwide who have to go through such experience and are dependent on others for their wellbeing and happiness. The encouraging news, however, is that the rapid advancement in technology has seen the innovation of better systems for assisting the disabled, including the blind, such as the AI glasses, which can provide intelligent navigation capabilities to the blind. This paper reviews the design of a smart cane, i.e., A smart stick for the blind, equipped with obstacle recognition using AI Technologies adds more virtual visibility in their journey. It shows that such a stick can be a significant boon to the blind.
In this paper, we focus on the recommender software. It uses artificialintelligence (AI) and operates once this anomaly detection software detects an anomaly [ 22 ]. The sensor data are coupled with the drone’s current direction obtained either via the drone’s on-board navigation system or from a compass mounted with the sensors and the obstacle detection data from the drone’s collision avoid- ance mechanism. The recommender uses these data as input to an off-policy deep learning model to recommend the direction of travel for the drone according to the current prevailing conditions, surroundings and sensor readings. This deep learning AI guides the drone to the site of the anomaly and the drone can transmit the exact anomaly site coordinates (and sensor data if needed) back to base for further investigation as appropriate. This is particularly important for safety-critical incidents or where the inves- tigators have to wear hazardous material suits or breathing apparatus with only a very limited time of usage (often around 20 minutes). They can head straight to the tagged location while the drone performs further sensor analyses. The AI recommender allows the human pilot and on-board collision avoidance or the drone’s autonomous navigation system (including collision avoidance) to focus on the actual navigation and the collision avoidance. This latter
Road safety is not an issue that will resolve itself, every citizen has responsibility for actions. We commit our combined knowledge, technology and network to promote safety. Between 25 percent and 33 percent of global crash are work related and 36 percent of occupational deaths worldwide are due to road crashes. Hence, crash avoidance system and devices help the driver and, increasingly, help the vehicle itself to avoid collision. This literature survey implement one such safety subsystem, Adaptive Cruise Control (ACC) using the ultrasonic sensors. This system uses an ultrasonic set up to allow the vehicle, to slow when approaching another vehicle or obstacle and accelerate again to the pre-set speed when traffic allows. ACC technology is widely regarded as a key component of any future generations of smart cars, as a form of artificialintelligence that may usefully be employed as a driving aid.
Early works on this topic motivated by the International Aerial Robotics Competition (IARC) Missions 3 and 4 (AUVSI Association, 2018) running, respectively, on years 1998 –2000 and 2001–2008 showed promising results using unmanned helicopters and computer vision. In Mission 3, the aerial robot had to detect and avoid obstacles, identify survivors, and recognize drum containers. The winning team from the Technical University of Berlin (Kondak & Remuß, 2007; Musial, Brandenburg, & Hommel, 2000) was able to perform the target identification and localization tasks; however, their helicopter did not fly near the debris, but rather flew high over the area (Greer, McKerrow, & Abrantes, 2002). In Mission 4, the aerial robot had to identify a particular building and deploy a rover to accomplish a task inside it. A team of the Georgia Institute of Technology won this challenge (Johnson, Mooney, & Christophersen, 2013; Rooz et al., 2009) by completing the entire mission. Working in topics that relate to our presented work, the same team also developed a helicopter system able to fly over an area and acquire an accurate three ‐dimensional (3D) reconstruction using a pan ‐tilt‐mounted laser range finder (LADAR or LIDAR) and explored the 3D obstacleavoidance problem in simulation (Geyer & Johnson, 2006). A comparison of the 3D reconstructions obtained by performing an overview flight and acquiring and processing data from either a LIDAR or a camera is discussed on the work by Leberl et al. (2010). The IARC Mission 5 (2009) proposed the challenge of autonomously exploring an indoor area with tight spaces while searching for a target object on a wall. Mission 5 was fully accomplished using a quadrotor drone and a low ‐weight LIDAR and a stereo camera system as main sensors by a team from the Massachusetts Institute of
with the accompanying video and YouTube list 7 in Fig. 10(c). The mentioned failure case for the autonomous flights is difficult to overcome. If the MAV would turn and face a large open space, the distance for EdgeStereo could be far enough to compromise the quality for the velocity estimate due to the small base line of the stereo camera. As we already observed in Fig. 6, this would cause the pocket drone to drift, which is problematic when near a wall/obstacle after the turn. If an obstacle is not in its FOV, the chances of collision significantly increases. This could be solved by merging the check and turn node of the FSM, so it will only stop turning at a significant clear path. Another solution is to add a lightweight short range sensor on the sides of the pocket drone, so it will detect immediately if the drone is flying close and aside an obstacle. The obstacleavoidance logic will need some additional work, however the experiments show that Edge-FS can be used in navigation overall. During the autonomous flight, the pocket drone was stabilizing itself using the velocity estimates of its forward camera alone.
Robot navigation problems can be generally classified as global or local, depending upon the environment surrounding the robot. In global navigation, the environment surrounding the robot is known and a path which avoids the obstacles is selected. In one example of the global navigation techniques, graphical maps which contain information about the obstacles are used to determine a desirable path. In local navigation, the environment surrounding the robot is unknown, or only partially known, and sensors have to be used to detect the obstacles and a collision avoidance system must be incorporated into the robot to avoid the obstacles. The artificial potential field approach is one of the well-known techniques which has been developed for this purpose. Krogh, for example, used a generalized potential field approach to obstacleavoidance. Kilm and Khosla used instead harmonic potential functions for obstacleavoidance. On the other hand, Krogh and Fang used the dynamic generation of sub goals using local feedback information.
Robots are designed to be controlled by a controller, computer or similar devices. Basically mobile robots require an operator’s vision and intelligence for guidance and navigation. The movement for the robot itself has many methods such as wheels, legs and many more. This movement method is to ensure the smoothness of moving in different type of surfaces.
1. Inability to plan under uncertainty of dynamic environments: Conventionally, global planners rely on a complete map of the environment in order to calculate the ideal and collision-free path between the starting point and the ending point prior to execution of the robot. The original plans of those conventional algorithms must be revised accordingly if a dynamic environment is encountered (Dijkstra, 1959; Hart et al., 1968). In practise, environment of robots often includes various hazard sources that robots must avoid, for example landmines, fire in rescue duty, and war enemies. Since it is impossible or expensive to acquire their accurate locations, decision- makers know only their action ranges in most cases (Zhang et al., 2013). Mobile robots must be able to evade both static and moving obstacles (Ferguson et al., 2006). Algorithms such as sampling-based methods (Khaksar et al., 2012) are not suitable for online planning when involving moving obstacles, due to the fact that these methods are designed based on a static environment model. These models are time- consuming when applied to a dynamic environment (involving interpolation cycle during each update, see (Huptych & Röck, 2015)). Therefore, classical path planning methods such as Visibility Graph (Lozano-Perez, 1987), Voronoi Diagrams (Leven & Sharir, 1987), Grids (Weigl et al., 1993), Cell Decomposition (Regli, 2007), Artificial Potential Field (Khatib, 1985), Rule Based methods (Fujimura, 1991) and Rules Learning techniques (Ibrahim & Fernandes, 2004) are not practical (Mohanty & Parhi, 2013). Occasionally, these algorithms are optimized to handle a specific problem at the expense of sacrificing the performance of other parameters such as increasing of the computational cost of the algorithm.
The global navigation problem deals with navigation on a larger scale in which the robot cannot observe the goal state from its initial position. Global approaches can be classified into three basic categories: The roadmap methods, the cell decomposition methods and the potential field methods . Roadmap methods extract a network representation of the environment and then apply graph search algorithms to find a path . Different Roadmap methods differ in the way they define the set of nodes, the set of pathways and the algorithm of finding feasible paths . Cell decomposition approach, is a graph technique widely used in both static and dynamic environments as the implementation is easier, accurate and easy to be updated . Potential field methods treat the robot, represented as a point in configuration space, as a particle under the influence of an artificial potential field . One shortcoming of this approach is that a robot can get stuck in a local minimum of the potential. A variety of approaches have been proposed for a robot to find its way out of these spots, including active search, backtracking, and random walks .
control method and adjusting the factors of repulsion potential field in real time . A proposed algorithm is developed based on new potential functions using the distances from obstacles, destination point and start point, while keeping the simplicity of traditional APF methods,. The algorithm uses the potential field values iteratively to find the optimum points in the workspace in order to form the path from start to destination . In Ref. , an improved artificial potential field based regression search (Improved APF-based RS) method is developed, which can obtain a global sub-optimal/optimal path efficiently without local minima and oscillations in complete known environment information. Potential functions are redefined to eliminate non-reachable and local minima problems, and utilize virtual local target for robot to escape oscillations. As the planned path by improved APF is not the shortest/approximate shortest trajectory, we develop a regression search (RS) method to optimize the planned path. The optimization path is calculated by connecting the sequential points, which are produced by improved APF. Amount of simulations demonstrate that the improved APF method very easily escape from local minima and oscillatory movements. An approach witch deals with the navigation of a mobile robot in an unknown environment is developed based on the Artificial Potential Field (APF) method in which the target creates a virtual potential that attracts the robot while obstacles create a virtual potential that repels the robot. A new form of repelling potential is proposed in order to reduce oscillations and to avoid conflicts when the target is close to obstacles. A rotational force is integrated as well, allowing for a smoother trajectory around the obstacles. The results of experiments show the effectiveness of the proposed approach . Using the mobile robot dynamic characteristics of the working environment, a model based on classic artificial potential field on the basis of considering dynamic model of velocity potential field is presented and simulation results show that this method can effectively improve the performance of path planning .
In recent years, topics related to robotics have become one of the researching fields. In the meantime, intelligent mobile robots have great acceptance, but the control and navigation of these devices are very difficult, and the lack of dealing with fixed obstacles and avoiding them, due to safe and secure routing, is the basic requirement of these systems. In this paper, the modified artificial potential field (APF) method is proposed for that robot avoids collision with fixed obstacles and reaches the target in an optimal path; using this algorithm, the robot can run to the target in optimal environments without any problems by avoiding obstacles, and also using this algorithm, unlike the APF algorithm, the robot does not get stuck in the local minimum. We are looking for an appropriate cost function, with restrictions that we have, and the goal is to avoid obstacles, achieve the target, and do not stop the robot in local minimum. The previous method, APF algorithm, has advantages, such as the use of a simple math model, which is easy to understand and implement. However, this algorithm has many drawbacks; the major drawback of this problem is at the local minimum and the inaccessibility of the target when the obstacles are in the vicinity of the target. Therefore, in order to obtain a better result and to improve the shortcomings of the APF algorithm, this algorithm needs to be improved. Here, the obstacleavoidance planning algorithm is proposed based on the improvement of the artificial potential field algorithm to solve this local minimum problem. In the end, simulation results are evaluated using MATLAB software. The simulation results show that the proposed method is superior to the existing solution.
As stated in section A.4, potential fields can suffer from local minima which results in the algorithm failing to find a path. Harmonic potential field method is also an artificial function based on harmonic functions, which overcomes the limitations of potential field methods. Harmonic functions are solutions to the Laplaces equation (eq. A.1), the so-called harmonic equations (hence the name harmonic potential fields). The most important property of harmonic functions is that they are free from local minima. The core idea of this method lies in creation of only one minimum in the working environment i.e, the global minimum which is represented by the goal. If the goal is represented by a global minimum and no other minimum exists in the environment then the robot will arrive at the goal location always. Harmonic potential fields provide a solution to this.
Propeller is the component which attached to the motor. The propeller used in aerial drone is like the RC airplane rather than the helicopter blades. This is due to size of the helicopter blade which is not suitable to a multi rotor aerial drone. The number of blades does not mean more thrust to the aerial drone. A smaller in diameter of propeller will produce a smaller inertia which means it would be easier to speed up or slow down; suitable to be use in acrobatic flight type. Propeller is design into two different kinds; a clockwise (CW) rotation and counter clockwise (CCW) rotation. It is important to know which motor attached to which type of propeller as it will provide a downward thrust to the aerial drone.
Fanti (2002) develops a real-time control strategy to avoid deadlocks, collisions and situations known as restricted deadlocks in a zonecontrol AGV system. Singh and Tiwari (2002) propose an intelligent agent framework to find a conflict-free shortest-time path for AGVs in a bi- or unidirectional guide path network. Yoo, Sim, Cao, and Park (2005) propose a simple and easily adaptable deadlock avoidance algorithm for AGV systems. Their algorithm uses the graph-theoretic approach. Sarker and Gurav (2005) present a bi-directional path flow layout and a routing algorithm that guarantee conflict-free, shortest-time routes for AGVs.
Nevertheless, in this study, we developed an advanced method for human behaviour recognition in video employing CNN, where ImageNet trains the network. In this study, the human behaviour has been divided into four categories: (1) actions, (2) activities, (3) interactions, and (4) group activity . Then we parsed human interactions into small actions through deep models. Moreover, besides the extraction of features and recognition of objects from videos in this proposed method, we also incorporated the potential of camera drone videos [32-34,5]. The use of drone vides is to find the human body. It has different challenges like motion blur, noise, and distance of drone from ground, and brightness. However, the potential of this study is to recognize the human body during disasters for rescue and emergency management. It will build a platform for future study on artificial geospatial intelligence (GeoAI) and smarter map. This study improved the performance of the previous algorithms, which was developed by Ravanbakhsh et al. (2015) ; Weinzaepfel et al. (2015), Shamsipour et al. (2017). In this study, we utilized image processing techniques on videos taken from Da-Jiang Innovations (DJI) drone. We collected drone videos as well as the ground platform videos. These videos were captured by cameras and then employed a pre-processing method to enhance the image resolution in order to achieve better information. A pre-trained CNN extracts the vectors with conceptual features that specify the image objects. In this method, SVM determines the relationships between the frame objects and labels the videos. However, this study contributes (a) a new method and algorithm coded for recognition of human-object interaction. (b) Improving the performance method of action recognition with 4.92% accuracy as compared to the existing method mentioned above. (c) Developing the pre-processing component enhances the image resolution for better information, and (d) created opportunities to develop GeoAI for the smarter map and disaster emergency as well as rescue applications beyond in future studies.
The FLC was designed using three different fuzzifiers (triangular, trapezoidal and Gaussian) to represent the sensor reading values so that they can be interpreted by the inference mechanism. Moreover, two different implication methods (Mamdani minimum and Mamdani Product implications) are used in the interpretation of the IF THEN rules in the rule-base. The number of control rules used in the design was 243
The above block diagram shows the overall co-ordination of the autonomous car. The structure consists of motors, relay circuits i.e. dual coil 12v dc relay board, IR sensors, Solar panel, 12v battery, 5v regulator. Whenever the 12v battery get discharged, the solar panel takes the thermal energy provided by the sun and charges the battery. As IR sensor and microcontroller requires 5v to operate, in order to convert the 12v into 5v,we require 5v regulator to regulate the 12v voltage. IR sensor is given as input to the microcontroller and it drives the DC motor using relay board.
A field experiment was performed during which a group of people were instructed to perform some obstacles avoidance tasks at two levels of normal and high speeds. Trajectories of the participants are extracted from the video recordings for the following intentions: (i) to find the impact of total speed (ii) to observe the impact of the speed on the movement direction, (iii) to find out the impact of speeds on the lateral direction. The results of the experiments could be used to enhance the current pedestrian simulation models.