Three aspects explain how to build a "three-dimensional world" for robots

From the undergraduate to postgraduate level, and after I started my business in 2014, I have been focusing on the field of vision for about 5 years. Therefore, it can be seen clearly that in the past few years, the perception-based algorithm with vision as the core, including face recognition, object recognition, spatial positioning, navigation obstacle avoidance, etc. is being based on the underlying artificial intelligence. The type of algorithmic architecture (such as machine learning, deep learning, and the recent intense learning enhancements) and the development of hardware sensors are undergoing rapid changes.

In the laboratory, my main research direction was to build a visual perception-based robot vision system for special machines, including micro-miniature drones and functional robots, which was developed after 2011. The advancement of a wave of perceptual layer algorithms and the innovation of sensor hardware have provided favorable support for such research, which has so far promoted the hot and sought after artificial intelligence projects in the entire capital market.

Below, I will tell you how we built a "three-dimensional world" for robots from three aspects: robot vision system, vision technology principle and future development trend.

We know that the term "robot" was proposed by a Czech writer in a philosophical drama in 1920. By 1950, another American writer Asimov systematically proposed "robotism". The concept and the famous three laws of robots are given. After that, from 1970, with the rise of computers, the development of modern control technology and sensor technology, robots began a process of real productization. It was also from that time that a robot based on a CCD chip camera could provide people with optical image information records at a certain time, which also formed the earliest robot vision system. It is worth mentioning that in 1969, the American Apollo moon landing spacecraft was equipped with a camera based on CCD sensor chip, which provided a systematic reference for the hardware architecture of the robot vision system. Due to the simple memory storage capacity, the robot at that time can perform simple repetitive tasks, but there is no perception and feedback control ability to the surrounding environment. We call the robots of the time the first generation robots.

Advancing into the 1980s, visual sensors, force tactile sensors, proximity sensors, and computers entered a period of rapid development during this period. In particular, the discovery of Moore's Law represented that the speed of information technology development has indeed reached its peak during this period. The robots in this period have already possessed certain perceptual capabilities, and can obtain part of the information of the working environment and the working objects, and perform certain real-time processing to guide the robots to perform the work. For example, in the figure below, we saw the Shakey mobile robot developed by the Stanford Research Institute in the United States. It has a sensing device such as an electronic camera and a range finder. It has established a layer-to-top layered control mechanism and the most advanced vision system at that time. Used to help robots perform independent reasoning, motion planning, and real-time control in an unstructured environment. This is one of the most mature achievements of artificial intelligence technology applied to mobile robots at that time. The birth of Shakey has also opened the curtain of intelligent mobile robot research.

Since then, countries around the world have begun to invest in the study of mobile robots, and in this, the visual system is recognized as the core entrance for robots to move toward intelligence. Because in the advancement of research, people need robots to have a better sense of the environment, logical thinking, decision-making ability, and even independent work according to operational requirements and environmental information. For example, the ALV autonomous vehicle researched by DARPA in the United States in the 1990s can choose road sign recognition to achieve navigation, achieve a mobile imaginary of 10km/h, and adopt advanced technology such as stereo vision and satellite navigation. In 2004, the Mars Exploration Opportunity and Courage developed by NASA successfully landed on the surface of Mars, carrying the most advanced image acquisition and stereo vision technology at that time, helping the detector to complete unknown tasks on the complex surface of the planet. It is also in such a period that the importance of 3D vision systems on mobile robots was first proposed.

Through the above introduction, it is not difficult to find that the robot vision system developed from the 1960s and 1970s is actually based on the most advanced algorithm technology and hardware sensors in different periods. In the algorithm technology of vision system, through the development of decades, four levels of user interaction, recognition perception, motion decision and data optimization are formed, which respectively realize somatosensory recognition, target following, human eye following; map construction, scene understanding , object recognition; positioning and positioning, autonomous navigation, path planning; image optimization, depth optimization, other data optimization, and many other well-known algorithms. On the hardware sensor, it is also divided into three levels: front-end sensor performance, integrated processing chip and embedded algorithm. In today's titanium confession sharing class, I will mainly introduce the main implementation principles of 3D vision.

In the previous introduction, we saw that the robot vision system has come along and is inseparable from the evolution of optical sensors. It can be said that the development history of the visual system for decades is the history of the evolution of optical sensors. Today, we generally classify the visual sensors carried in the robot vision system into three categories: one-dimensional line array sensors represented by single-line laser radar, two-dimensional area array sensors represented by embedded cameras, and special light sources. 3D depth sensor. Among them, the three-dimensional depth sensor represented by special light source is the most important and most important sensor to realize the three-dimensional vision system of the robot. The acquisition quality of the three-dimensional data directly affects the algorithm result and decision control of the back end of the mobile robot.

At present, the mainstream technology for realizing three-dimensional depth sensors is generally developed after 2010. The routes have the following types: based on monocular structured optical technology routes, based on binocular structured optical technology routes, and based on flight time method technical routes. In November of this year, the M-32 three-dimensional sensor launched by the service-oriented robot manufacturer is based on the binocular structure light principle and integrates the visual sensor of the embedded vision algorithm. The principle of structured light is to use the optical diffraction principle of the laser to project a specific pattern through the sensor to accelerate or assist in the acquisition of the depth map. The specific patterns can be divided into rules, pseudo-random or random point speckles and special pattern spots. The advantages are high precision and fast refresh rate, but the disadvantage is that it is not suitable for use in outdoor bright light environment. The principle of time of flight, also known as the TOF principle, is the depth acquisition achieved by the principle that the modulated light source receives different phases at different distances and inversely calculates the distance according to the propagation speed of the light. The advantage of this principle is that the measurement accuracy does not vary with distance. Decrement, but the disadvantage is low resolution and large environmental disturbances.

NB ASSET GPS TRACKERS

NB asset GPS trackers,NB-IOT asset GPS trackers,4G Asset GPS Trackers,Asset Tracking Device,LTE Waterproof Asset Trackers

eSky wireless Inc , https://www.eskygpsiot.com