When can a sports robot have a sports brain, it is no longer an "artificial mental retardation"

Running, jumping, and backflips on the lawn, one pit one step at a time; while climbing the stairs, the shaking stairs keep trembling...

These "trick-or-treating behaviors" all come from Boston-powered robots.

Since the company released its first robot motion video, almost every time, their new developments have attracted great attention from the industry. There is no other reason. The robots marked with the "Boston Dynamics" label are all top in the industry in terms of motion control.

But even if the movement robot of Boston Dynamics realizes the actions that children can do easily, the performance is still unsatisfactory. For the robotics industry, why is athletic ability so difficult?

This has to start with the development of motion robots.

In fact, it is difficult to trace the accurate development history of sports robots. The compass car invented by the ancient Chinese more than two thousand years ago can probably be regarded as the originator of the concept of moving robots. The car is equipped with a "mechanical transmission system", which uses the differential information of the two wheels to indicate the direction when the steering is transmitted. This is undoubtedly a successful application of mechanical control technology.

By 1961, Unimation had developed the world's first industrial robot for General Motors, which was used to produce car doors, window handles, shift knobs, lamp fixtures, and other hardware inside the car. So far, robots have begun to replace humans and have been gradually used in many traditional industries.

With the development of modern related technologies and the emergence of innovative applications, pure mechanical motion has long been unable to meet the demand. The concept of motion robots has gradually been popularized, and the difficulties in research and development have also been widely valued.

The motion robot should be a comprehensive hardware system integrating environment perception, dynamic decision-making and planning, behavior control and execution and other functions.

To achieve this, we must gather a number of cutting-edge technologies such as sensor technology, information processing, electronic engineering, computer engineering, automation control engineering, and artificial intelligence in the "body" of the robot. Robots are required to have a flexible "motion brain". However, none of these can be achieved in a short time in the world.

"For the robotics industry, there are three key technologies: computer vision, semantic understanding, and motion control. Motion control is a more difficult level." When talking about the key technologies of robot control, Vincross COO Xu Kaiqiang once said.

How difficult is robot motion control? Are they called "mentally retarded" too wronged?

It is not difficult to see that the so-called "motion robots" are far from expectations in terms of behavior control or environmental judgment. To put it bluntly, they are still in the stage of "artificial mental retardation".

For this reason, Zhang Rui, the founder and CEO of Iron Man Robot, believes, "They need to have a'sports brain', that is, a'sports brain'."

Looking at the various sports robots currently launched, either can only "stand" in place and use a screen with a "painted" face to say the preface and no words, and you can nod and shake your head a little; or they are very mechanical in every move. , Swinging his hands and feet to dance, there is no practical application meaning at all.

The ability to independently plan routes, sense the environment and move automatically while avoiding obstacles, and be able to move without mechanical movement through flexible joints is the most basic requirement for a robot "human".

Among them, there are two major technical difficulties: mobility and activities.

If you don't move well, you will bump into it randomly; if you don't move well, you will have no spirituality, just like mental retardation.

The key to movement is the "chassis", and the mechanical structure design, sensors and algorithms are indispensable.

According to the way of movement, mobile robots can be roughly divided into three categories: wheeled, crawler and footed. The key to the movement of the first two types of robots is the lower body, which is commonly known as the "chassis."

Yu Yuanlong, Dean of Cobos Robotics (Nanjing) Artificial Intelligence Research Institute, said that the "chassis" is actually just a carrier, which carries and applies three key technologies: sensors, intelligent algorithms and driving mechanisms.

Judging from the current working status of mobile robots, its main workflow should be "sensors are used to perceive the environment, and the environmental information is transmitted to the "control brain", that is, intelligent algorithms; the "brain" gives "execution" by analyzing environmental information Command', ultimately controls the output behavior of the drive mechanism."

In terms of sensors, more accurate and high-precision environmental perception is the most important indicator for evaluating its performance. But high-performance sensors mean high costs. Taking Cobos as an example, to overcome the problems of sensor dependence on imports and high cost, owning the sensor patents independently researched and developed is the only solution.

And intelligent algorithms have also been the focus of major robot manufacturers in recent years. Among them, it involves many key technologies such as environment perception, navigation and positioning, obstacle avoidance, decision-making, vision, voice, human-computer interaction and so on.

At the same time, Dean Yu also said, “At present, various manufacturers are still in the shallow water area in terms of intelligent algorithms, but have achieved some basic functional applications. Simply put, the current robots are not smart enough.”

The "not smart enough" here is embodied in three aspects: first, the perception ability is not smart enough; second, the behavioral decision-making ability is not enough; third, the human-computer interaction is poor.

As far as the current development trend is concerned, the combination of multi-modal sensors and intelligent algorithms is an effective solution. When robots can effectively perceive environmental changes, accurately reason and make decisions, and provide targeted services based on user needs, it is time for us to enter the "deep water zone".

In terms of driving structure, it involves many mechanical engineering issues such as mechanical structure design, effective driving force, reducer and so on. These are directly related to the flexibility of the robot when it moves.

Presumably, no one wants to see the scene where the robot either does not move, or rushes out "swoosh" and knocks down on the wall.

Because of this, most of the current mobile robot applications are limited to indoor scenes. The robots needed by the market should have more abilities, such as doing physical movements naturally, being able to "pour tea and water" or even "cross mountains and mountains".

This involves the robot's ability to move.

When it comes to "activity" robots, the most common ones are stage performance robots.

Spinning, waving, kicking...These are the most common movements of dancing robots. Among them, the best-known robots that have spent the Spring Festival Gala above are the best-known. And their biggest advantage lies in the technology-independent research and development of servo steering gear and gait algorithm.

Zhou Jian, the founder and CEO of Ubican, once said, “Servo servos are a difficult technical barrier. Especially high-torque servo servos are a big problem in the robotics industry globally.”

Compared with ordinary motors, the servo steering gear also covers a variety of technical issues such as deceleration, sensing, motors, chips, improved algorithms, and rotation degrees of freedom. Because of this, the servo steering gear can greatly improve the joint sensitivity of the robot when it moves. Therefore, Ubican did not give up on the development of servo steering gear, even though they spent nearly 5 years and half a billion.

In fact, servo steering gear is not a new concept. Japan, Switzerland and other major robotics research and development countries have long had "servo steering gear" related products. It's just that the import cost was extremely high at that time, even the lowest cost price was $50 per piece, and the early demand was not high, so it has not been widely used.

In this regard, Zhou Jian said: “Generally speaking, the cost of the servo servo alone can account for half of the total cost of the robot. In this way, once the quantity is small, the supply chain will not start at all. The corresponding chips and CPUs , The cost of the battery can't be brought down either."

However, as users have increased the sensitivity of robot activities, servo servos have long become the key to differentiated competition among robot manufacturers. The original “low demand” has become a false proposition.

However, as Zhou Jian said, self-research is a way out for small and medium-sized enterprises in the early stage of "small volume" and high import costs.

So if the robot can move, how can we make it move?

For example, when you are in a hurry, the normal physiological response of "humans" will cause us to take large steps, and the walking speed and body swing will change accordingly; and if you have more time, you may be He wandered in small steps, his head swayed and his arms swayed at random.

Robots should also be like this. In different scenarios, gait and each part of the body will show different movement states. In layman's terms, the robot should be more down-to-earth when moving, instead of repeating a single or a few actions dumbly.

This requires an equally grounded algorithm.

When does it move? Where to move? How to move? How often? What's the magnitude? These are all things that must be considered when doing the algorithm. Especially when the scene is diversified and complex, the amount of engineering for algorithm development will increase exponentially.

Xu Kaiqiang also said, "It seems simple to let the robot move its hands and feet, but in order to do this well, it took us three full years."

And Zhang Rui said, "If people can live in this world widely, then robots should be able to be widely used in various fields, which requires different algorithms to give them different functions."

And this actually has a solution. As long as you add a Gazebo-based simulation platform and an Ubuntu-based ROS system to the robot, it will be able to access more developers in terms of hardware interface and open source.

Iron Man did just that. "What we provide is not only a hardware product, but also a software development platform for developers."

Coincidentally, Vincross also starts from motion control technology to provide a convenient development platform for robot developers.

"It is recognized in the industry that robots will become a new generation of computing platforms, just like mobile phones ten years ago. But in fact, due to the high barriers to robot development technology, most developers are still unable to perform robot development tasks. Therefore, we The need for a platform that can provide convenient operations is a prerequisite for development and innovation."

Simply put, it is to build a simple development platform based on the robot to attract developers to give the robot more functions according to their own ideas.

For example, let the robot "fight the bull": Based on the real-time image processing function of the camera that the robot has, write the corresponding code so that it "rushes" out immediately when it receives the red signal.

Of course, this is just a simple game application. With the increase in the number of sports robot scenes and the increase in application innovation, the open platform will greatly shorten the development cycle and increase the speed of industrial development.

Although the ability is limited, the blue ocean in the vertical field has long appeared

According to incomplete statistics, there are more than 40 companies specializing in the research and development of sports robots in domestic startups alone. Among them, more than 80% have been commercialized in the market, and they are distributed in various fields.

Last year, Suzhou InterContinental Hotel used mobile robots to replace humans in leading the way and delivering meals. In its 48 reviews based on the Ctrip platform in December 2017, 11 mentioned the keyword "robot".

Through calculation, the average score of these 11 comments is 4.98 points, while the average score of other keywords without "robot" is only 4.72 points.

“Many of our customers had a comprehensive score of only 4.5 points before applying robots, but after applying robots, the scores have increased greatly, and some even rose to more than 4.8 points in a short period of time. In addition, 1 robot It can replace up to 4 human workers, and in terms of labor costs, it has also contributed to the hotel." said Zhang Mingju, senior vice president of Yunji Technology.

In addition to hotels, in industry, such as Jingdong’s storage robots; in household services, such as Cobos’s sweeping robots, have also been widely recognized by users. In more vertical fields such as agriculture, medical care, urban security, national defense, and space exploration, mobile robots have also begun to go to dangerous areas that are difficult for humans to reach. They are engaged in some extremely challenging tasks, and the value they play is increasing. Come bigger.

There is no doubt that the blue ocean in the motion robot industry has emerged.

Although Boston Dynamics Robotics is an "industry legend", it still hasn't reached a satisfactory level.

But what needs to be clear is that among the three key elements of robots, motion control is indeed the most difficult and complex, and it cannot be accomplished overnight.

However, it is worth mentioning that the development of current robot application scenarios is not yet mature, and most of them are fixed and extremely vertical, such as hotels, shopping malls, and experimental fields. Although the requirements for professionalism are relatively high, the processing and complexity requirements for robot motion are not required. High, some even as long as they have the ability to move the chassis.

This is the achievement of the robotics industry.

Of course, more interactive and personalized robots are ultimately what users need. In the future, robots will provide more and more services, and truly "human-like" general-purpose robots may also become the mainstream. Although the road ahead is long, it is still worth fighting for.