"Put three fried chicken on the right side of the cabbage" Mitsubishi Electric is a new technology that facilitates the introduction of robots

On February 28, 2022, Mitsubishi Electric announced the development of "teaching-less robot system technology" that facilitates the introduction of robots. Incorporating technologies such as voice recognition AI that understands the speaker's intention with high accuracy and automatic optimization of articulated robot movement, even workers without specialized knowledge can easily automatically generate robot movement programs and are equivalent to humans. It is said that the work speed of can be realized.

According to the company, it is the first industrial robot manufacturer to develop voice-based work instruction technology. Toru Oka, director of the Advanced Technology Research Institute of Mitsubishi Electric, said about the technology, "The main target is the food and logistics fields, and the food factories where the menu changes frequently, etc. Can contribute to the promotion of automation in the work process. " It is scheduled to be put into practical use in 2023.

Features of "Teachingless Robot System Technology" developed by Mitsubishi Electric

Mr. Toru Oka, Director of Mitsubishi Electric Advanced Technology Research Institute, and Mr. Masato Seki, Manager of Autonomous Control System Development Project Group, Advanced Technology Research Institute of Mitsubishi Electric

Robot startup time reduced to "1/10 or less" without programming

The teachingless robot system technology announced this time has two major features: "program generation / adjustment facilitation technology" and "automatic robot motion optimization technology". As a result, the start-up adjustment and operation of the robot can be performed more easily than before, and the work speed can be increased to the extent that it is comparable to human workers.

The food manufacturing industry is mentioned as one of the main targets of application because the robot utilization rate in this field is still low. Mr. Masato Seki, Group Manager, Autonomous Control System Development Project, Mitsubishi Electric Advanced Technology Research Institute, said, "In recent years, the food manufacturing industry has been receiving increasing demand for automation due to labor shortages." Point out that there is.

"In the background (low utilization rate in the food field), it is difficult for beginners to perform preparatory work such as robot operation and program creation, and it takes time to adjust the system startup and switch types (such as changing the menu of bento boxes). There are reasons such as the fact that human work is often faster even if the system is built. "(Mr. Seki)

This technology will eliminate these barriers to introduction. The specific system is an arm robot and a controller that moves it, three vision sensors that scan moving objects / recognize irregular objects / static environment scans, a force sensor that controls the robot hand, and a personal computer that controls these. It is composed. In addition, worker instructions can be entered using a 3D sensor camera and tablet.

For example, in a lunch production line, a 3D sensor (or a camera at the tip of an arm robot) scans the area around the robot from various angles in advance to generate a 3D model in virtual space. Understand the work area of ​​the arm robot, the position of objects, obstacles, etc. After that, using the vision sensor that scans moving objects, the lunch box flowing on the line is accurately grasped, the arm robot is operated correctly in real time according to the work instructions, and the prepared foods are packed.

Schematic block diagram of teachingless robot system technology

Work instructions to the robot can be given by voice or simple item selection using a tablet. In the case of voice input, you can say something like "pack three fried chicken in the first (section) of the lunch box." Also, when instructing by entering items, all you have to do is touch and select items such as "what", "where", and "how many" on the tablet. In addition, since the trajectory of the arm robot's operation according to these instructions can be AR-displayed / confirmed as an image on the tablet, accidents due to unexpected operations can be prevented.

In the past, it was necessary to control all the movements of these robots by programming, so it took time to achieve the optimum movements. By using this teaching-less technology, it is possible to automatically create a robot motion program with simple instructions, just like teaching a new task to a human worker. According to the company, the work time required for program generation / adjustment can be reduced to "less than one-tenth of the conventional time".

In addition to increasing the recognition rate by adopting voice recognition technology that is resistant to noise around the work environment, by adopting the "intention understanding AI model", the work content that the worker instructed in natural words can be performed with high accuracy. Can be estimated. Ambiguous instructions such as "a little more right" and relative positions such as "right of cabbage" are also possible. Mitsubishi Electric's AI technology "Maisart" is characterized by its miniaturization, and it can be installed in edge devices such as tablets by taking advantage of this.

Adopting a voice recognition / touch interface that even field workers can easily handle, anyone can easily teach the robot.

AI cancels noise and realizes highly accurate voice recognition.In addition, we have adopted "intention understanding AI" that allows workers to give instructions in natural words.

Automatically optimizes arm and hand movements, enabling faster work

Another feature of this technology is "automatic optimization of robot movement". For example, just by specifying the start point and end point of the operation, a trajectory that does not interfere with the surrounding equipment is automatically generated at the second level without conducting a test run. At this time, the trajectory that the maximum movement amount axis can maintain the maximum speed is calculated under the constraint of speed and allowable torque, and the optimum acceleration / deceleration pattern and the movement of the hand that performs stable grip while minimizing the stop time are also automatically calculated. Therefore, the operating time of the robot is shortened.

In addition, it is equipped with a "grasping recognition AI" that reproduces the bulk state of irregular objects by simulation and uses a trained model to infer the gripping position by hand at high speed and with high accuracy. Even for ingredients that do not have the same shape, such as fried chicken, the angle of the arm and the opening width of the hand can be determined from the image, and the deformation of the ingredients and the balance due to the gripping position are taken into consideration for stable gripping. You can work at high speed. With these technologies, it is said that the work speed equivalent to that of human work has been achieved.

"The work time for the robot to grasp the member and place it in the specified place is the shortest of 2 seconds per pick, which is equivalent to that of a human. Conventionally, it took 3 seconds or more, which is one-third of the work time. We have been able to reduce it "(Mr. Seki)

"Grip recognition AI" that infers the position to be gripped at high speed and with high accuracy

In addition, we have developed a "ROS-Edgecross linkage function" that enables easy monitoring and analysis of the entire production line due to the connectivity and multi-vendor characteristics of Edgecross, which contributes to productivity improvement and quality improvement. Can be done. Mr. Seki explained, "By collecting all system logs including ROS (Robot Operating System) with Edgecross, line performance evaluation can be facilitated."

Overview of ROS-Edgecross linkage function

Mitsubishi Electric plans to put this technology to practical use in the fields of food and logistics, and to expand it to other fields such as electrical and electronic fields. "In the future, we will consider the possibility of expanding into new markets such as supermarkets and convenience stores where articulated robots have never been used," said Mr. Seki.

■ Related sites