"put three fried chicken nuggets on the right side of the cabbage" Mitsubishi Electric facilitates the introduction of new robot technology.

Mitsubishi Electric has announced the development of "teaching robot system technology" to facilitate the introduction of robots in food manufacturing, logistics and other fields. Through the technologies of high-precision speech recognition AI to understand the speaker's intention and automatic optimization of the action of multi-joint robot, even the operator without professional knowledge can easily generate the robot action program automatically and achieve the same working speed as human. [see more photos]

Photo: Asky

　Mitsubishi Electric announced the development of "Teaching Lever system Technology" on February 28, 2022 to facilitate the introduction of robots. Including high-precision speech recognition AI to understand the speaker's intention, automatic motion optimization of multi-joint robot and other technologies, even operators without professional knowledge can easily and automatically generate robot action programs to achieve the same working speed as human beings. This is the first time an industrial robot manufacturer has developed a voice-based job instruction technology, the company said. Gancher, director of the Mitsubishi Electric Advanced Technology Comprehensive Research Institute, explained on the technology: "Food factories with food and logistics as the main objectives and frequent menu switching can contribute to the automation of food loading and sorting processes that have previously been difficult to introduce robots." It is expected to be put into practical application in 2023. There is no need for programming, and the start-up time of the robot is "less than 1/10". The teaching robot system technology published this time has two characteristics: "program generation / adjustment capacity facilitation technology" and "automatic optimization technology of robot action". As a result, the start-up adjustment and operation of the robot can be carried out more simply than before, and the operation can be made at a high speed as high as that of human operators. As one of the main goals of the applicable object, the food manufacturing industry is listed because the utilization rate of robots in this field is still very low. Guan Zhengren, manager of the autonomous control system development project team of Mitsubishi Electric Advanced Technology Comprehensive Research Institute, pointed out that "due to the shortage of manpower in the food manufacturing industry in recent years, the demand for automation is getting higher and higher." at the same time, there is also a large threshold for introduction. In the context of "(the utilization rate in the food field is low), the preparation of robot operation and program production is very difficult for beginners, and it takes time for system startup adjustment and variety switching (easy menu changes, etc.). There are also a lot of situations in which people who finally build the system work faster. There is such a reason "(Mr. Guan) to remove such barriers to introduction is this time's technology." The specific system is the arm robot and its controller, moving object scanning / amorphous object recognition / static environment scanning for three vision sensors, robot hand control force sensor, these control computer composition. In addition, the operator's work instructions can be entered using three-dimensional sensor cameras and tablet computers. For example, in the production line of lunch box, a three-dimensional sensor (or a camera at the front of the arm robot) is used to scan the perimeter of the robot from various angles to generate a three-dimensional model in the virtual space. Master the working area of the arm robot, the position of the object, obstacles and so on. Then, the vision sensor is used to scan the moving object, accurately grasp the lunch box flowing on the line, operate the arm robot in real time and correctly according to the work instruction, and load the home cooking. Using a tablet, you can indicate the work of the robot by voice or simple project selection. In the case of voice input, it can be said that "put 3 fried chicken nuggets in No. 1 (compartment) of the lunch box" and so on. In addition, in the case of instructions through project input, just touch and select "what", "where", "a few" and other items on the tablet. In addition, according to these instructions, the trajectory of the arm robot can be displayed / confirmed by AR as an image on the flat panel, so it can also prevent accidents caused by unexpected actions. In the past, because it needed to be programmed to control all the movements of such a robot, it took time before performing the best action. By using this non-teaching technology, just like teaching human workers a new job, the robot's action program can be made automatically with simple instructions. According to the company, the working time spent on program generation / adjustment can be reduced to "less than 1/10 of the original". In addition, in addition to using the speech recognition technology which is also very noisy around the working environment to improve the recognition rate, by using the "intention to understand AI model", the content of the assignment indicated by the operator in natural language can be deduced with high precision. It can also be specified in vague instructions such as "a little more right" and relative positions such as "right of cabbage". Mitsubishi Electric's AI technology "Maisart (Maisart)" is characterized by miniaturization, which can also be installed on edge devices such as tablets. Another feature of this technology is "automatic optimization of robot movements", which can automatically optimize the movements of arms and hands and achieve high-speed work. For example, as long as you indicate the starting point and end point of the operation, you can automatically generate a track in seconds that does not interfere with the surrounding equipment, without trial operation and so on. At this time, under the limitation of speed and allowable torque, it is calculated that the maximum axis of movement can maintain the trajectory of the highest speed, and the action of the hand which is stably controlled while minimizing the optimal acceleration and deceleration mode and stop time is also calculated automatically, thus shortening the action time of the robot. In addition, by simulating and reproducing the bulk state of atypical objects, the "control recognition AI" of the training model is used to infer the control position of the hand with high speed and high precision. Even if the shape of the component is uneven, such as frying, to determine from the angle of the arm in the image and the opening width of the hand, further stable grip taking into account the balance due to deformation and clamping position of the fixture, high-speed work is possible. Through this technology, the operation speed is the same as that of human work. "the robot controls the component and places it at a designated location for a working time to achieve a minimum of 2 seconds for each pickup equal to the human hand. In the past, it took more than 3 seconds to reduce working time by 1/3 "(Guan's), coupled with the development of" ROS-Edgecross joint function ". Through the characteristic connectivity and multi-supplier nature of Edgecross, the entire production line can be easily monitored and analyzed, contributing to productivity and quality improvement. Guan explained: "it is convenient to evaluate line performance by collecting system-wide logs, including ROS (Robot Operating System), through Edgecross." Mitsubishi Electric's policy is not only to make the technology practical in the fields of food and logistics, but also to extend it to other fields such as electrical machinery and electronics. "in the future, we will also study the possibility of developing into new markets, such as supermarkets and convenience stores where multi-joint robots have not been used before" (Mr. Guan). Edited by Otsuka / TECH.ASCII.jp, Wendahara