Patent Portfolio

Patent No. 5892531 ／ 6529039

Link Sequence Mapping Apparatus, Link Sequence Mapping Method, and Program

Expiration 2035.11.16 ／ 2036.03.18

【What it can do】Human movements captured using motion capture can be directly applied to robots with fewer arm joints, as well as to animal-shaped robots such as dogs, birds, and snakes. This eliminates the need to write conversion formulas for each robot, thus reducing the adjustment costs associated with creating new robots. Even if a person adopts an unexpected posture, the system can still find a solution, preventing the robot from becoming immobile in live events or customer service settings.

【How it works】The model's link sequence takes coordinates from multiple points, and the link sequence is calculated to minimize the objective function based on the distance to the corresponding points on the robot, thereby outputting the joint angles. The key point is to solve this as an optimization problem each time, rather than using a transformation formula.

USE CASESCharacter/animal-shaped robots, remotely controlled avatars, exhibition displays, rehabilitation support equipment

Patent No. 6347335

Robot System, Positional Relationship Acquisition Apparatus, Positional Relationship Acquisition Method, and Program

Expiration 2036.12.07

【What it can do】The assembled robot will automatically determine the order of its joints and its ground contact status. This eliminates the need for manual input of a correspondence table showing which motor rotates which joint, preventing malfunctions due to input errors. This is especially effective in development environments where multiple prototypes are being assembled and modified.

【How it works】Angular velocity sensors are placed on each rotating axis, and each axis is rotated one at a time. When one axis is rotated, the other axes are divided into two groups with different angular velocities, so it can be determined that the rotated axis is between these two groups. If the angular velocity of the other axes is small when the tip axis is rotated, it is the free end; if it is large, it is the grounded end.

USE CASESRobot kits, educational materials, self-diagnosis of multi-joint mechanisms, assembly inspection.

Patent No. 6703679　＋ USA 12067494 ／ China ZL2020800030176

Counting device, learning device manufacturing device, counting method, learning device manufacturing method, and program

Expiration 2039.02.01

【What it can do】Even if identical objects are piled up, densely packed, or overlapping in a photograph, the system can determine the number of items from a single image. It can directly replace processes that were previously done manually, such as inspection, inventory, and counting individual fish in aquaculture and seedling rearing. It can also count objects of sizes it hasn't studied, making it suitable for environments where objects of different sizes are mixed together.

【How it works】The model is trained using pairs of captured images and images in which a "training figure" smaller than the object is placed at the position of each object. During inference, images appear where the objects have been transformed into smaller figures, and these figures are counted. A confidence level is assigned to each figure based on its proximity to the training figure, and only those that exceed a threshold are counted.

USE CASESInspection/inventory of parts and pharmaceuticals, counting fry (young fish) or livestock, livestock head counts, event crowd estimation.

Patent No. 6742623　＋ USA 12705896 ／ China ZL2020800927711

Monitoring device, monitoring method, and program

Expiration 2039.11.13　共同出願：アイシースクウェアパートナーズ株式会社

【What it can do】Even if you increase the number of surveillance cameras, you don't need to increase the number of people monitoring the footage. Since there's no need to create separate detection models for each installation location (e.g., for shoplifting, for falls), the cost of multi-site deployment is significantly reduced. The detection models can be updated later, and the same system can be used not only with fixed cameras but also with vehicles, drones, and robots.

【How it works】The AI ​​determines the type of object being monitored from the video footage, and then retrieves an anomaly detection model from the cloud that is appropriate for that object. The device itself only stores the framework for the detection process; the actual content is changed to suit the specific situation on site.

USE CASESMulti-store retail monitoring, facility surveillance, infrastructure inspection, mobile security robots, wide-area surveillance via vehicles or drones.

Patent No. 7412826

Image compositing apparatus, image compositing method, and program

Expiration 2043.07.28

【What it can do】This learning screen overlays example footage with the user's own footage, solving the problem of "not being able to tell which is me." The more misaligned the images are, the easier it becomes to distinguish between them, making it particularly effective for beginners. It can also be integrated into existing video learning systems by simply changing the display method.

【How it works】The blending ratio of the two images is continuously and repeatedly changed over time (including a mode that moves back and forth in a sinusoidal wave). Control is also defined so that the further you are from the example, the darker your own image becomes.

USE CASESRehabilitation motion training, dance/sports form correction, manufacturing work instruction.

Patent No. 7432275

Video Display Device, Video Display Method, and Program

Expiration 2043.07.28

【What it can do】Even if the example is a first-person video from a head-mounted camera, learners can compare it using just one handheld camera. Since learners do not need to prepare the same equipment and setup environment, those who create the teaching materials only need to distribute them, and there are no barriers to adoption for those receiving them. This makes remote skill transfer a viable business model.

【How it works】The system recognizes the learner's skeletal structure from the video footage and moves a 3D object corresponding to the example in accordance with the learner's movements. The video is generated and arranged so that the relative positions of the example camera and subject match the relative positions of the viewpoint from which the 3D object is viewed.

USE CASESRemote skill transmission, online practical training, deployment of standardized work procedures across multiple sites.

Patent No. 7486860

Image compositing apparatus, image compositing method, and program

Expiration 2043.08.04

【What it can do】This system not only shows learners the discrepancies between their actions and the example, but also provides guidance on how much and in which direction to move to achieve the correct result. Since learners can progress independently without an instructor present, it reduces the need for teaching staff. It is particularly suitable for tasks involving the use of tools and equipment, where physical movement alone is insufficient for accurate coordination.

【How it works】The system uses feature points from the example video to determine the positional relationship between the example and its camera, and from the user's own video to determine the positional relationship between the target object and its camera. From these two, the system calculates the positional relationship between the example and the target object. Based on this, a tracking guide is generated and displayed together with the composite video.

USE CASESSelf-study of techniques using equipment/instruments, tool operation training, skill certification tests, practical exam preparation.

Patent No. 6671672

Dialogue device, speech generation method, and program

Expiration 2039.05.13

【What it can do】This allows you to construct reception and guidance dialogues without relying on pre-defined scenarios. It's based on a patent for speech generation combining deep learning and knowledge systems (ontologies), and forms the foundation of the dialogue engine "Dialog AI".

USE CASESReception/guidance robots, audio guidance systems in facilities, customer service assistance.

Patent No. 7152821

Mobile Body, Mobile Body Control Method, and Program

Expiration 2042.04.28

【What it can do】Without the need for weight sensors, acceleration and deceleration can be automatically adjusted according to the weight of the loaded cargo. This reduces product damage and rework due to cargo shifting, and eliminates the need to over-adjust the speed setting. Since it doesn't add sensors or wiring, it can be implemented on existing machines with just a modification to the control software.

【How it works】After starting to move, the system accelerates at a predetermined rate, and the weight of the object being transported is determined from the motor current value at that time. Based on the obtained weight, the movement mechanism is controlled to prevent the load from shifting.

USE CASESTowing/transport in logistics centers, factory parts delivery, retail backroom stocking.

Patent No. 7481029

Mobile Body and Program

Expiration 2042.05.12

【What it can do】This prevents misfollowing, such as accidentally following people or furniture. It ensures that the robot doesn't mistakenly follow the wrong person or object, even in busy corridors or warehouses where workers in similar attire are present. The patent covers not only transport but also robots used for guidance, security, surveillance, and entertainment.

【How it works】The system identifies the target object using camera images and determines its position relative to the user from the distance measurement results corresponding to that area. Furthermore, while moving, objects whose position changes less than other objects in the user's coordinate system—that is, objects moving together—are considered the target object to be tracked.

USE CASESGuide robots for warehouses, airports, and hospitals with high foot traffic; patrol security.

Patent No. 7649571

Mobile Body and Program

Expiration 2043.11.10

【What it can do】Even if it loses sight of its target, it won't get stuck. Transport will continue uninterrupted even on layouts with many turns and intersections, eliminating the need for a person to go and call it back. Since the route it can run is predetermined, its recovery movements can be predicted, and safety explanations become easier.

【How it works】At the start of movement, the camera and range sensor determine the position of the target being tracked, and the range sensor is used during movement to store the latest position in memory. If the position can no longer be determined, the robot selects one of several pre-set routes in its own coordinate system that corresponds to the last known position and follows that route.

USE CASESWarehouses with complex aisle layouts, retail backrooms with heavy foot traffic.

Patent No. 7649580

Mobile Body and Program

Expiration 2044.05.13

【What it can do】The need to stop the robot, load the items, and start it again after each pick is eliminated. The worker simply walks towards the robot, and the robot follows, reducing the time required for each round. Operational inconveniences such as the robot stopping too close to the wall and being unable to get close to the worker are also resolved.

【How it works】It will have a follow mode that tracks people and a non-follow mode that does not track. While running in non-follow mode, if the relative positions of the tracked object and the moving object become constant, it will automatically switch to follow mode.

USE CASESWarehouse picking operations, combining fixed route transport between processes with human-following tasks.