State-of-the-Art Mobile Intelligence: Enabling Robots to Move Like Humans by Estimating Mobility with Artificial Intelligence

Xue-Bo Jin, Ting-Li Su, Jian-Lei Kong, Yu-Ting Bai, Bei-Bei Miao, Chao Dou
2018 Applied Sciences  
Mobility is a significant robotic task. It is the most important function when robotics is applied to domains such as autonomous cars, home service robots, and autonomous underwater vehicles. Despite extensive research on this topic, robots still suffer from difficulties when moving in complex environments, especially in practical applications. Therefore, the ability to have enough intelligence while moving is a key issue for the success of robots. Researchers have proposed a variety of methods
more » ... and algorithms, including navigation and tracking. To help readers swiftly understand the recent advances in methodology and algorithms for robot movement, we present this survey, which provides a detailed review of the existing methods of navigation and tracking. In particular, this survey features a relation-based architecture that enables readers to easily grasp the key points of mobile intelligence. We first outline the key problems in robot systems and point out the relationship among robotics, navigation, and tracking. We then illustrate navigation using different sensors and the fusion methods and detail the state estimation and tracking models for target maneuvering. Finally, we address several issues of deep learning as well as the mobile intelligence of robots as suggested future research topics. The contributions of this survey are threefold. First, we review the literature of navigation according to the applied sensors and fusion method. Second, we detail the models for target maneuvering and the existing tracking based on estimation, such as the Kalman filter and its series developed form, according to their model-construction mechanisms: linear, nonlinear, and non-Gaussian white noise. Third, we illustrate the artificial intelligence approach-especially deep learning methods-and discuss its combination with the estimation method. it is in relation to the surrounding environment. When we walk on the street, we can see that a car is moving and that the leaves on a tree are moving because of wind. Fortunately, we know that the house will not move. Although we see the house moving backward as our forward movement proceeds, we also know that the house is not moving. According to our position relative to that of the house, we also can judge our speed and that of the car, predict the location of the car, and determine whether we need to avoid the car. For robots, the so-called mobile intelligence means that they can move in the same manner as human beings and exercise the same judgment. A robot on a road, for example, an autonomous car, needs to answer three questions constantly during movement: where am I, where am I going, and how do I go? It first needs to judge its movement relative to all other kinds of movement (cars, trees, and houses in the field of vision) and then has to avoid other cars. Among these tasks, the ability to understand surroundings is critical. The performance of an autonomous car heavily depends on the accuracy and reliability of its environmental perception technologies, including self-localization and perception of obstacles. For humans, these actions and judgments are dominated by human intelligence, but for machines, they are extremely complex processes. We first define robotic mobile intelligence. When moving, a robot should have two basic capabilities: (1) to localize itself and navigate; and (2) to understand the environment. The core technologies enabling these capabilities are denoted as simultaneous localization and mapping (SLAM), tracking, and navigation. SLAM constructs or updates the map of an unknown environment while simultaneously keeping track of the robot's location. It is a complex pipeline that consists of many computation-intensive stages, each performing a unique task. Tracking refers to the automatic estimation of the trajectory of a target as it moves. The tracking algorithms are based mainly on estimation theory. Here, the target is a general one, either the robot itself or others. When the robot wants to "track itself", which means that the robot wants to know its own motion, called "navigation", by which the robot determines its own position and direction of movement. We now define foreground and background targets. We usually set the target to track as a foreground target, such as the cars. The house does not need to be tracked, nor do other stationary objects, so it is set as the background. Some mobile aspects that do not need to be tracked, such as the movement of leaves, are called background targets. The most difficult aspect of video tracking is to distinguish the difference between foreground and background targets. In current navigation systems, researchers are trying to replicate such a complicated process by using a variety of sensors in combination with information processing. For example, vision sensors are commonly used sensors in robotic systems. In the video from a camera fixed on a robot moving on the street, the houses are moving, the cars are moving, and even the trees on the roof have their own motion because of the robot's own movement. How to obtain the mobile relationship between the robot and foreground target from so many moving targets has always been an area of intense research interest. The optical flow method [1] is a widely used method to determine this relationship, which assumes that the apparent velocity of the brightness pattern varies smoothly almost everywhere in an image. Researchers have done much useful work in this area [2, 3] . The disadvantage of the optical flow method is the large computational overhead. Moreover, it is difficult to find the correct optical flow mode when the robot is moving at a high speed. To overcome such a difficulty inherent in a visual sensor, other sensors have been introduced, for example, inertial measurement units (IMUs) and global positioning systems (GPSs), into the navigation system. GPSs are now widely used in cars [4] . In most cases, GPSs provide accuracy within approximately 10-30 m. For autonomous cars, however, this accuracy is insufficient. IMUs have become smaller, lower cost, and consume less power thanks to miniaturization technologies, such as micro-electro-mechanical systems or nano-electro-mechanical systems. Because the measurements of the IMUs have unknown drift, it is difficult to use the acceleration and orientation measurements, to obtain the current position of a pedestrian in an indoor navigation system [5] .
doi:10.3390/app8030379 fatcat:zx2u5ox4ivcvtm2vb2yg2kmbqm