"The industrial robots are so stupid now."

of computer history

"Robots will change the world!" "The future belongs to the robot!" "Recently, some Silicon Valley investment agencies, including Google Venture, have repeatedly reiterated such comments, but the road to universal access to robotics is not smooth sailing." If the development of the robot in the computer history, the robot's hardware itself has developed to a certain degree of maturity, however, there is no cross-platform programming language to enable developers to easily programming for robots. Mujin, the founder of the industrial robot programming platform from CMU and MIT, hopes to clear the barrier and has recently received 6 million dollars in funding from Japanese venture capital Jafco and Utec.

Robots are not obedient to 1-year-old children.

In fact, the robots we call are not "humans", but some machines that mimic human behavior, and under the engineer's "set-up" they can walk and carry objects. However, "teaching" a robot to walk is far more difficult than teaching a 1-year-old child to walk, because the robot's "brain" is blank, its hands to lift the foot should be at what angle, to what height, all need engineers with logic and experience one by one set. Not only that, as the "public misconceptions about robotics," this article says, because there is no mature and universal platform for robot behavior complex behavior programming, robot engineers need not only good programming skills, the underlying hardware must have a considerable understanding. As a result, this has long been a huge technical threshold for robot programmers and the scarcity of relevant talent, so the robot is currently only active in a few areas to complete simple and repetitive action.

The product line is a field where more robots are deployed. Tesla's production lines, for example, and a portion of Foxconn's production lines have deployed a large number of robots to collaborate on specific frequencies to repeat certain actions. Compared with human workers, robots do not need to rest, nor because of fatigue and reduce efficiency, or produce errors, in many ways than the human assembly workers have a better advantage.

Can industrial robots be easily tuned? naive!

But despite the good prospects, the robot's current use of assembly lines is still very limited, in addition to the above mentioned, the robot engineers of the huge technical barriers caused by the scarcity of talent, as well as the high cost of employing scarce talent caused by the enormous human costs, the production line This particular scenario also brings many additional problems

One of the most obvious problems at the moment is the motion planning of robots. Assembly space is a complex environment, how the robotic arm works together, how to avoid obstacles, and how to collaborate quickly and efficiently is not a simple problem. The movements of some of our seemingly commonplace robots are often the work of engineers who have been out of tune for many nights. The Japanese company, Mujin, tries to solve the problem once and for all through its programming platform for industrial robots.

General purpose robot programming platform Mujin

Mujin as a general-purpose programming platform for industrial robots has actually done two things, general and motion planning. Using the Mujin programming platform, process Master can directly control the different manufacturers of the mechanical arm planning behavior, and no longer need to learn the programming language or the underlying hardware logic, reduce learning costs; The second is that this is not only an advanced programming language platform to compile more abstract language into machine language, It is also an advanced robot control platform, which can automatically plan the robot's motion trajectory after inputting the abstract behavior.

The figure above is the code on the four line Mujin platform, and the following figure is the action of the compiled manipulator

In fact, the trajectory planning of a robot is far from being as easy as it sounds. In the limited space, how to avoid other obstacles, how to make the robot arm to move the shortest distance, how to make the mechanical arm wear smaller, need more than a few experienced engineers a few days more months of debugging. And the movement of the manipulator is often counter-intuitive, in order to achieve a better trajectory, the robotic arm will often make 180° rotation and other human actions can not be realized.

Mujin in robot motion planning is a popular in the current academic circle called RRT (rapidly exploring Random tree) algorithm, simply in the three-dimensional space, random generation can be from point A to B path, and the shortest shortest path can be realized by the robot from the machine. James Kuffner, Mujin's Ctorosen Diankov, is one of the inventors of the RRT algorithm. Therefore, when the three-dimensional model of the assembly shop is imported into the Mujin platform, any motion instruction of the manipulator is input, the Mujin can calculate the superior trajectory directly, and then tell the manipulator how to move. Furthermore, the algorithm can calculate the trajectory of a multidimensional manipulator, such as the motion of a 7-joint manipulator.

Mujin's Vision

Although the RRT algorithm sounds "violent", it is the most efficient and versatile trajectory algorithm known to man, and the efficiency of the algorithm depends on the user's understanding, optimization, and implementation. Mujin, director of software development at MIT, told me that Mujin is now almost able to do real-time trajectory planning, that is, almost at the same time as the input command, the robot can show the calculated trajectory. Such a computational efficiency is at a leading level that few other university labs or commercial companies can achieve.

Mujin's team is composed of programmers from China, Japan, the United States, Ukraine, India, and many other countries, and the office is located in the world's most widely used industrial robots in Japan. The Mujin platform has been deployed in a number of production lines, including Canon, Honda, Nissan and other Japanese companies.

At present, a production line with 20 general robotic arms costs more than $40~100 for the picker, but the human cost of manipulating the picker is often several times the cost of the picker. Therefore, ideally, using a small number of engineers plus the Mujin platform can save a lot of money for a production line, and because the trajectory of the algorithm is often faster and more efficient than that of the engineers ' experience, the productivity of the Mujin will be improved. However, this is only the ideal situation, the stability of the premise of the pursuit of efficient big manufacturers Why trust a start-up? In this regard, Liu Huan confidently told me, "I have some doubts, but they (the representative of Japanese companies) is really very experienced, see the Mujin platform to calculate the trajectory of the movement to know it to do faster and more efficient." "Mujin also gained 6 million U.S. dollars in funding from two Japanese venture firms from Jafco and Utec.

In the longer term, Mujin hopes to extend the platform to more areas. Portable programming language, and intelligent trajectory planning is almost the robot in the road to popular popularity must be two-step road. By providing advanced, universal, no need to consider the motion planning of the robot programming platform, so that developers and even ordinary people can better use the robot, to achieve "industrial for everyone" is Mujin's mission and vision for the future.

For many tech optimists, Mujin is doing something exciting, but the road is not only unusually cumbersome but difficult in reality. But I think Mujin is making a very good attempt: in the integration of academia and the industry, human beings from the popular use of the robot is a step away from the unknown.