The Massachusetts Institute of Technology (MIT) has beenrobotA pioneer in scientific and technological research, this laboratory has developed world-shaking military Robots such as Cheetah and Atlas. So, with the development of cutting-edge artificial intelligence technologies such as DeepMindAlphaGo and Atlas, what new trends will emerge in robotics research? At the CCF-GAIR Global Artificial Intelligence and Robotics Summit Robotics Session, the director of MIT Robotics Laboratory, IEEE, AAAIFellow, and Daniela Rus, academician of the National Academy of Engineering, gave a speech on this topic, and described the twelve cutting-edge technology trends in the field of robotics in the world:
“Moore’s Law” in Robotics
Maybe everyone thinks this picture is futuristic, but in fact we have realized it to a certain extent. Robots can be used in scenarios such as delivering packages, cleaning the environment, sorting goods, autonomous driving, and living assistance. In addition, we have also seen some companies, they have invented two kinds of single-arm collaborative robots, and used in production.
What these examples tell us is that robots have indeed moved from sci-fi to current scientific reality. We can make robots more capable and more intelligent.
It is worth mentioning that there are also subversive laws similar to “Moore’s Law” in the field of robotics. Including manufacturing tools, design tools and other fields, the rate of disruption of internet performance changes every 6 years. Likewise, the number of factory robots doubles every five years. At present, we have temporarily verified this fact, and I believe that the frequency of this disruption will be higher in the future.
In the future world, everyone may have a robot. Robots are as common as cars running on the road. I call it a world of “ubiquitous robots”.
These robots will be able to perform many tasks in collaboration with humans. Of course, now we have not reached such a stage, because there are still many technical problems to be solved. For example, how do robots interact with people, how do they reason and solve problems on their own… and how can we make new robots quickly and cheaply?
Next, I will share with you some technology trends, which can help us solve the above problems.
Twelve Robotics Trends
Previous robots were all steel bodies, but such structures did not adapt well to various environments. Soft body refers to making the structure of the robot soft and flexible, like the structure of the human body. Generally speaking, the body structure of the soft robot is made of soft silicone, which enhances its adaptability and can adapt to different unknown environments.
Based on the principle of muscle operation, we found that such a configuration makes the robot more agile and able to complete certain tasks more quickly. In addition to soft silicone, we can also drive the structure of the soft body with water or air. For example, this (enlarged version) robot, which looks like a snake, and these bubbles on the surface can drive the robot’s activities by zooming in and out.
We can see that when the robot is placed in the pipe, it can automatically detect the surrounding environment, and the adaptability of the plastic type is unmatched by the steel body robot.
In the same way, we can also create Robotic fish. It is mobile like an actual fish, capable of 90-degree turns and can quickly evade predators. Thanks to its soft-bodied tail, the robotic fish can move up and down the water.
We have seen the importance of soft bodies, and a new journal, Soft Robotics, has been out for two years. Through this journal, we know that the importance of soft robots in the robotics subject ranks the highest, which means that everyone pays the highest attention to soft robots.
Manipopulation: flexible operation
In addition to soft robots, another technology that improves and enhances robots is: flexible grasping and handling operations.
A steel-bodied robot can only see the size of the object clearly, aiming at where each finger is placed to grab the object, but this is not how humans operate. When we want to take something, we reach out and grab in a very continuous motion, without thinking about the size or which finger to use. It is precisely because of the precise requirements for the position of the fingers that the robot’s grasping behavior has great limitations, and they have no way to deal with irregular objects.
And soft handling came into being. Because there is no need to look carefully at where the object is placed, and it is not controlled by the shape. For example, it can grab eggs and paper strips. Because this robot has a very flexible structure, it can freely deal with various uncertainties.
We can also give robots the ability to recognize actual objects by embedding some simple sensors. Of course, this cannot be done 100%, and the recognition accuracy in some scenarios is low. Horizontal grabbing or using two fingers will have a higher success rate, because horizontal grabbing accumulates more data and knows how to grab it; while two-point grabbing has less information.
Even with soft structures, sometimes robots fail. why? If the robot can’t catch it, it can tell the human what’s wrong, but it can’t.
It can be found by observation that when a robot performs a task, a little human intervention can completely change its plan. How to improve the interaction between human and machine? If the robot could simply say “Help me, I’m stuck,” that would solve the problem, but it can’t yet. In addition, if the robot can also introspect and calculate new decision-making actions based on its own data, this failure can be avoided.
So we want to give robots this ability. We have developed a program planning system through which the robot can think about its own course of action – when it gets stuck, it can think “why it got stuck, how can I get out of this obstacle”, or communicate this idea to a human- “Please move the table up”.
So imagine that the robot has to have the ability to communicate, to communicate with the outside world very clearly and clearly. Otherwise, it can only say “help me”, and humans have to check to see what is wrong with it when they come over, so the efficiency is very low.
Cloud big data helps learning
We know that robots also need to learn. However, we humans can accept a large amount of data every day from birth to learn, and for robots, data storage is easy to run out of memory. A self-driving car has 1TB of data in an hour, which is difficult to analyze. Therefore, we need to improve the degree of abstraction, so that the collected data can reach a higher level and reduce the storage pressure and calculation amount.
For example, on the left is a GPS data stream. If we can build a meaningful structure for this GPS data stream, we can deduce some extractable information from it, and then do high-level reasoning. For example, when the autopilot reaches a certain location, it knows what task to perform.
Extract data from the data stream, perform abstract processing, and summarize meaningful information – this is the core-level technology to be shared next – through an algorithm, analyze some small data sets in big data, these small data The set can reflect the result of the entire data operation.
The same example: We use Coresete’s method to obtain the number of datasets by analyzing the video, and then focus on different colors, from which more and more complex videos can be analyzed. There are 16500 frames in the movie screen, we only need to use 1152 Coresete data points to start the analysis.
When there is only one robot, the tasks that can be completed are limited, and we need many robots to form oneautomationsystem. So, the fifth trend is multi-robot systems.
When several robots are grouped together, each robot has its own work. Of course, if you are building a cabin now, one robot will be responsible for moving parts, and another robot will be responsible for other tasks. So, we can see that these four robots are collaborating.
Robots must be able to communicate and coordinate with each other in order to know when to cooperate with their peers to perform tasks. This is a challenge. They need to know both their own mission and the overall collective mission.
Manufacturing on demand
Our goal is to allow a robot to pass3D printingIt is printed directly by the machine, but this is not an ordinary shell printing. In a 3D printer, there must be a driving mechanism, and we can see the electronic structure inside. This is actually a very complex mechanism.
Empower everyone to design their own robot. Is this idea crazy? With the basis of database, programming tools, 3D printing and other technologies, although not all robots can be completed automatically, it is true that many steps can be completed automatically.
At the same time, robots are actually very versatile and can penetrate into every aspect of our lives. For example, if a micro-object is swallowed by mistake, we can make a micro-folding robot, send it into the intestine, and let it wrap up the foreign body and take it out of the body through the folding type, thus helping us to avoid minimally invasive surgery; Alternatively, microrobots could be used to provide stomach treatments to humans.
Learning Robots Early
Let students start robotics learning early, using programming tools to create a variety of robots. What we hope to achieve is to attract students with the magic of robots – not only the robot shell, but also soft skills such as learning programming. Let students and children enter the world of robots happily, and gradually invest in the Robot Industry.
academia andindustryCooperative competition in the world
We are now facing an unprecedented change in the computer industry. We need forward-looking or whimsical ideas from academia, and cooperation from industry to turn these ideas into products. At the same time, the government should also get involved and come up with the right implementation plan so that the robot can really play a role. US DARPA is a good example.
A year ago, Toyota approached MIT. The company says that driving is fraught with dangers and that 1.5 million people are now killed in road accidents every year around the world. Through industry-university cooperation, academic wisdom can be applied to the difficulties of the industry, so as to better develop intelligent driving vehicles.
MIT is also conducting research and development on intelligent driving vehicles, and Singapore proposed such a plan in mid-2010. The combination of smart driving and autonomous ride-hailing services provides cities with a network of autonomous vehicles.
However, in general, we can only drive in simple environments, and there are still many obstacles to overcome in real driving environments.
Business Investment and Entrepreneurship
Now the world is also waking up to the opportunities in robotics. In recent years, there has been a large-scale investment in the field of robotics, and in 2015, there were about $2 billion in investment and mergers and acquisitions.
Innovation in China
China really wants to lead change in robotics. I put forward a vision here. In the future, there will be more robots working side by side with workers, and robots in the future will be more advanced than today.
In fact, do we need to worry about robots replacing us? In fact, we should be more worried that we are not building robots fast enough. In China, only 20% of people will still be of working age by 2050, so the production of robots should be accelerated to make up for the labor shortage.
(Original title: Twelve cutting-edge technology trends that must be known in the field of robotics)