NEC develops robot control AI that flexibly responds to changes in warehouse operations and layout- "World Model" applied to significantly reduce learning time and promote the introduction of robots -
Tokyo, March 3, 2023 - NEC Corporation (NEC; TSE: 6701) has developed robot control AI that can flexibly respond to frequent changes in operations and layouts in distribution warehouses, factories and other locations. By utilizing this technology, NEC will promote the introduction of robots to worksites that are conventionally dependent on human labor, and contribute to the improvement of productivity and working styles.
In recent years, due to labor shortages and other factors, the introduction of automated equipment, such as robots, is progressing for simple operations that repeat the same work. However, work operations require the handling of a wide variety of shapes and arrangements, and it may take several months for AI to thoroughly learn the work in advance. Moreover, items that are being handled and work layouts may change frequently. For this reason, the introduction of robots is limited to certain work operations, and many sites rely on human labor.
By applying a “World Model" to robot control though, NEC has significantly reduced the time required for pre-learning and flexible responses to changes in environmental conditions, such as differences in equipment and work layouts. Going forward, NEC will proceed with verification of this technology, aiming for practical use during fiscal 2024.
About World Models and robot control AI applications
Recently, the research and development of World Models is attracting global attention in the field of machine learning in order to promote the use of robots. A World Model is a technology that enables predictions to be made about what will happen in the real world as a result of certain actions without actually trying them. For example, if a person opens their hand while holding an item, the item will drop to the floor due to gravity. Consequently, depending on the posture of the item when the hand's grip is released, the item will fall, etc., and come to a rest. Based on common sense accumulated through experiences in the past, humans can imagine what will happen in the future as a result of their actions and act appropriately. However, in order to achieve the same results with a robot, it is necessary to comprehensively program this common sense, which is a major issue in the autonomous control of robots. World Models are expected to be a key technology in the provision of autonomous control, in which robots understand the structure and common sense of the real world and decide their actions while imagining the future.
According to NEC research, this is the world's first robot control AI that applies a World Model and autonomously generates and executes optimal operations with minimal failures, even under previously untested work conditions. This ensures that even irregularly placed items that are different sizes and shapes than those learned can be accurately grasped and placed in the correct position and orientation.
Features of this technology
- ①Produces a high success rate according to working conditions
In order for a robot to perform operations such as arranging and packing various kinds of randomly placed items into boxes, it is necessary to generate motion candidates, such as the grasping and placing of items in designated places, so that the robot does not drop the items. For such tasks, conventional machine learning methods use robot control rules to generate motion candidates according to learned work conditions. As a result, inaccurate motion candidates may be generated and executed depending on the actual work conditions. For this reason, the work success rate is only about 70% according to NEC research.
In addition to learning through robot control laws, NEC's new technology learns through a “Motion Prediction Model," which is a World Model that predicts the success or failure of a task. Just before an operation is executed, a Motion Prediction Model is used to predict the success or failure of multiple motion candidates generated according to the characteristics of the work conditions, then the operation candidate with a high success rate is selected and executed. NEC's simulations have confirmed a work success rate of approximately 95%, which is a level of stability that according to NEC research can be utilized in the field, and can be achieved without comprehensively learning the working conditions that may occur in the field.
- ②Pre-learning in a short period of time
In the past, in order to learn operations that can handle a variety of items and tasks, it was necessary to thoroughly learn the expected arrangement of items as well as the work patterns. For example, to learn the task of placing rectangular items of various sizes in predetermined positions and orientations in a box, it was assumed to take several months or more to pre-learn the tens of thousands of patterns that are required.
In contrast, NEC has shortened pre-learning in two ways. First, by introducing a Motion Prediction Model, the need for comprehensive learning has become unnecessary. Second, by introducing an active learning method it has become possible to learn a Motion Prediction Model and robot control rules with a smaller number of patterns. Specifically, in order to generate operations that can handle the above-mentioned work of placing rectangular items in a box, it is possible to shorten the pre-learning time to several days.
NEC announced part of this technology at the International Conference on Intelligent Robots and Systems (IROS 2022), a top international conference in the field of robotics, held in Kyoto, Japan, from October 23-27, 2022.
About NEC Corporation
NEC Corporation has established itself as a leader in the integration of IT and network technologies while promoting the brand statement of “Orchestrating a brighter world.” NEC enables businesses and communities to adapt to rapid changes taking place in both society and the market as it provides for the social values of safety, security, fairness and efficiency to promote a more sustainable world where everyone has the chance to reach their full potential. For more information, visit NEC at https://www.nec.com.
NEC is a registered trademark of NEC Corporation. All Rights Reserved. Other product or service marks mentioned herein are the trademarks of their respective owners. © NEC Corporation.