Kawasaki Robotics //treeforeurope.com/dk/ Thu, 26 Jun 2025 06:48:56 +0000 da-DK hourly 1 //www.altis-dxp.com/?v=6.8.4 //treeforeurope.com/tachyon/sites/35/2022/02/cropped-site-icon.png?fit=32%2C32 Kawasaki Robotics //treeforeurope.com/dk/ 32 32 Kawasaki Robotics //treeforeurope.com/dk/blog/human-ingenuity-kippei-matsudas-journey-from-nfl-ai-competition-winner-to-ai-solution-developer-at-kawasaki-robotics/ Fri, 28 Feb 2025 05:58:56 +0000 urn:uuid:fedd10f7-d10c-49c3-9b09-908ab0526095

Kippei Matsuda: Robot Technology Development Department, System Technology Development Center, Technology Development Division. Dr. (ingeniør)

Beskriv NFL-udfordringen, løsningen, og hvad der gjorde den spændende?

Konkurrencen havde til formål at identificere spillerkollisioner, og mere end 1.000 dataforskere verden over konkurrerede om at bygge en nøjagtig løsning ved hjælp af optagelser fra NFL-kampe og sensorinformation. Amerikansk fodbold er kendt for at være en af de hårdeste og mest fysisk krævende sportsgrene i verden med meget kontakt, og selv om spillerne bærer beskyttelsesudstyr, er skader hyppige. Især hovedkollisioner resulterer ofte i alvorlige skader eller handicap, og det har været en udfordring at finde måder at reducere virkningen af kollisioner på sammen med effektiv behandling, efter at en skade er opstået. Hvis vi nøjagtigt kunne identificere, hvilke spillere der havde slået hovedet i løbet af en kamp, kunne vi give effektiv behandling og fremme forskningen i hjelmens virkning og i, hvordan man kan mindske påvirkningen. At udføre denne type forskning manuelt ville være ekstremt tidskrævende, så NFL afholdt denne konkurrence for at bruge AI-teknologi til at løse dette problem.

Nøglen til at vinde denne konkurrence var den vellykkede integration af to forskellige datatyper, video og sensorinformation. Vi brugte videobilleder til at identificere spillerkollisioner og estimerede spillerpositioner ved hjælp af sensorer, der var fastgjort til spillerne. Udviklingen af et AI-system, der omhyggeligt analyserede og forudsagde selv de mindste ændringer i spillernes position og kropsholdning, såsom at sidde på hug eller falde, skilte os ud. Denne innovation forbedrede vores nøjagtighed betydeligt i forhold til andre deltagere. Resultatet var, at vores behandlingshastighed var 83 gange hurtigere end manuel betjening, og opgaver, der tog 3-4 dage, kunne nu klares på bare 2 timer, hvilket vi satte stor pris på.

Kippei Matsuda
Generelt handler AI-billedanalyse om at opdage objekter i billeder, men i denne konkurrence skulle vi tage højde for spillernes tredimensionelle (3D) positioner på banen på skærmen, hvilket var en spændende udfordring, som vi aldrig havde prøvet før. Det er svært at analysere data, fordi man er nødt til at se på dem igen og igen, men NFL-videoen var så stærk, at jeg nød at se den igen og igen. Takket være dette var jeg ved konkurrencens afslutning i stand til at forestille mig spillernes bevægelser bare ved at se på videoernes titler.

Hvad fik dig til at deltage i NFL-udfordringen, og hvad var resultatet?

Det hele startede med, at jeg tænkte, at det kunne være en del af mine studier. Teoretiske aspekter kan læres fra bøger og andre kilder, men det er svært at vide, hvordan man bruger faktiske data og kører en simulation. Som udvikler er det vigtigt at røre ved og analysere data og gentage forsøg og fejl. Jeg blev tiltrukket af konkurrencen, fordi den indeholdt materialer, der førte til praktisk læring.

“At deltage i konkurrencen gav mig mulighed for at omsætte AI-udvikling til praksis”

Kippei Matsuda
Ærlig talt følte jeg aldrig, at selve konkurrencen var svær; det, der var svært, var at finde tid til mine studier og min familie, da jeg arbejdede på det i min fritid. Når jeg legede med mine børn i parken, tænkte jeg pludselig: “Måske kunne jeg gøre den del på denne måde”, og det generede mig. Det var svært at slappe af. Jeg sad i toget og rystede. Fra starten af projektet troede jeg ikke, at jeg kunne vinde, men da jeg gjorde, var jeg begejstret. Jeg klager ikke over at have vundet, og jeg var glad for at dele nyheden med alle omkring mig. Jeg kunne næsten ikke få noget arbejde gjort den dag!

Hvordan arbejder du med AI-vision i dag hos Kawasaki Robotics?

Jeg er i øjeblikket involveret i at udvikle produkter, der udnytter AI-vision specifikt til robotteknologi. AI analyserer billeder, der er optaget af kameraer, og behandler dem på forskellige måder; i vores depalleteringsløsning behandler den f.eks. billeder af det produkt, der skal læsses af. Depalleteringsløsninger bruges til at forbedre aflæsningseffektiviteten på distributionscentre og fabrikker. Depalleteringsløsningen er udstyret med 3D AI-vision og er i stand til at foretage meget sofistikerede analyser af det gods, den håndterer.
Kawasaki Robotics’ depalleteringsløsning er udstyret med 3D AI-vision og er i stand til at foretage meget avancerede analyser af det gods, den håndterer. Ved at specialisere os i aflæsning har vi opnået høj ydeevne til en lav pris.Kippei Matsuda
Manuel losning er meget kostbar og tidskrævende, og konventionelle robotløsninger er ikke fleksible nok til at håndtere arbejdet. I den henseende er det lykkedes vores depalleteringsløsning at øge nøjagtigheden og hastigheden af automatiserede aflæsningsoperationer sammenlignet med konventionelle løsninger.
Med konventionelle robotsystemløsninger skal alle pakkestørrelser og -former f.eks. registreres, før de plukkes. Hvis pakker med uregistrerede former kommer ind i workflowet, kan de ikke behandles. Vores depalleteringsløsning kræver dog kun, at de mindste og største størrelser registreres, og alle pakker kan behandles. I konventionelle robotsystemer skal robotten kende den korrekte størrelse og form på det produkt, der skal håndteres. Et kamera er nødvendigt for at genkende og bekræfte produktets størrelse og form. Hvis det bekræftes, at det er korrekt, vil robotarmen plukke produktet, og det er en tidskrævende proces at lære systemet det.
Vores depalleteringsløsning kræver ingen forudgående produktregistrering ud over minimums- og maksimumsdimensionerne, hvilket reducerer undervisningstiden betydeligt.KippeiMatsuda
Hovedårsagen til, at Kawasaki Robotics’ depalleteringsløsning har været en succes, er, at vi parrer billig kamerahardware med vores robuste AI-software til at behandle komplekse produktbilleder uden en masse ekstraudstyr. Vi opnåede en højtydende og brugervenlig løsning ved at fokusere på at udvikle et depalleteringssystem, og ved at indsnævre funktionerne opnåede vi en bedre omkostningseffektivitet end andre virksomheders produkter, hvilket gør det lettere for virksomheder at introducere produktet.
Himekawa, leder af produktudviklingen (tilhører General-Purpose System Section 2, General-Purpose System Department, Robot Division)

Hvordan vil depalletering og AI-løsninger fortsætte med at udvikle sig?

Jeg tror, at effektiviteten af depalleteringsløsninger vil stige, efterhånden som der indsamles flere og flere data, og AI-læringen skrider frem. For eksempel var en af vanskelighederne ved at udvikle en depalleteringsløsning snorene og båndene på overfladen af pakkerne. Når AI ser dem, kan den forveksle snorene med grænserne for pappet. Den kan så beslutte, at kassen er mindre, end den faktisk er, og tage den ved en fejltagelse. Men hvis systemet bruges forskellige steder i fremtiden, og der indsamles data, vil det være i stand til at lære flere variationer af pakker. Så selv om en kasse har snore, tape eller klistermærker, kan robotten aflæsse den korrekt baseret på dens tidligere erfaringer uden at blive vildledt. Det bliver nødvendigt at indsamle en masse data af god kvalitet til robotten. Jeg tror også, at AI, kaldet infrastrukturmodellen, bliver nøglen. Et eksempel er ChatGPT, som har tiltrukket sig meget opmærksomhed i de senere år. Da den underliggende model er trænet på baggrund af en stor mængde data, har den en meget høj genkendelsesevne, en slags sund fornuft. Grundlæggende modeller kan håndtere en række forskellige oplysninger, herunder tekst, billeder og lyd, og har potentiale til at udvide brugen af robotter dramatisk, ikke kun inden for logistik. I fremtiden vil jeg gerne udvide brugen af AI ved at gøre god brug af data og den grundlæggende model.

Klik her for produktsiden for Depalletizing Solution (japansk)

]]> Kawasaki Robotics //treeforeurope.com/dk/blog/story_1/ Mon, 06 Dec 2021 12:00:00 +0000 urn:uuid:9e8e0fb9-da44-43c4-a0bd-dc5d9d953808 Illustration af industrirobotter: duAro og R-serien But when it comes to actually implementing a robot, a lot of questions arise. “What tasks can you leave to the robot?” “What steps do I need to take?” “Who should I talk to in the first place?” Here are some of the best ways to take industrial robots from concept to reality.

Consult with Professionals

You’re starting at square one: No robots have been introduced yet. You are interested in automation, but not sure which tasks to automate. In the beginning, it can be difficult to judge whether a particular task can, or should, be automated or not. In this situation, there are resources to give you peace of mind when introducing a new robot system. In many cases, a specialized engineering firm called Robotic System Integrator (robot SIer) is responsible for planning, designing, and deploying robotic systems. In general, the SIer interacts between the user and the robot manufacturer, acting as a connection between them, and leads the way to the system installation. Kawasaki is a valuable robot maker that can even play the role of a robot system integrator.
Illustration af en situation, hvor du er interesseret i automatisering, men ikke ved, hvilke opgaver du skal automatisere.
Let’s take a look at the flow from the introduction of robots to installation. (An example of a basic flow is shown in Fig. 1.) In the early stages, the system integrator will conduct preliminary meetings and field observations to gain a better understanding. It’s important to know what the end-users are looking for and what’s going on in the industry, as well as basic requirements like budget, schedule, cycle time requirements, specs, variety, workspace, etc. Building a robotic system is a collaborative effort between the end-user, system integrator, and robot manufacturer, and a thorough understanding of the requirements is essential for success.

Create a robot-conducive environment

Next, you need to understand whether or not automation should be used. Robots are better at some tasks than others. For example, it is easy to accomplish repetitive tasks that require high degrees of accuracy, or dull, dirty, and dangerous jobs that aren’t ideal for humans to execute. But when it comes to complex applications requiring human senses such as sight, delicate touch, smell, and taste, extra equipment and sensors may be required, which can make a system more complicated and expensive. It is important to consider whether a robot is really suitable for the process you’re thinking of automating and whether a robot can demonstrate its power, productivity, and cost-effectiveness. Even in the common case of partial improvement of production processes, the first step is to sort out the tasks that should be performed by humans and the processes that should be performed by robots, taking into account the above viewpoints.

After narrowing down the processes to be automated by the robot, work elements are disassembled in a way that makes sense for the robot. For example, a person might think of a task the following way: Remove the screw and place it on the product on the jig. When you finish tightening, put the finished product in the next box. But in the case of a robot, it is necessary to subdivide each task:
Step 1: Remove the screw
Step 2: Place the product on the jig
Step 3: Place screw in the designated location
Step 4: Tighten screw
Step 5: Pick the finished product
Step 6: Place finished product in box

En illustration af, hvordan man opdeler arbejdselementerne, så de er lette for en robot at forstå
At this point, it’s easy to overlook the details of the human worker’s movements. For example, turning the parts inside out when placing them on a rack, visually checking the product for foreign objects, or tapping the surface to check the sealing performance are all simple but important movements. Once a robot is made to do the detailed work that the workers are doing almost subconsciously, it is necessary to construct a system not only for programming each and every operation but also for linking the tools used and the processes before and after.

At the same time, it is essential to create an environment in which robots can operate.
For example, if there is no storage space for equipment needed before and after the automated process, there is no problem in the operating space of the robot itself, but the process could be delayed. It is very important to design while imagining not only an automated process but also keeping in mind a realistic process flow such as whether this equipment is smoothly linked to the tasks that come before and directly follow that process. This requires a macroscopic perspective with a bird’s-eye view of how to smoothly pass complex elements such as workers, robots, parts, products, space, and time from upstream to downstream.

Et makroniveaudiagram, der viser, hvordan man problemfrit kan overføre komplekse elementer såsom arbejdere, robotter, dele, produkter, rum og tid fra opstrøms til nedstrøms.
Figure 1. Robot introduction process chart

Follow Up After Install

Once the details of the system are established from start to finish, a risk assessment is conducted based on the basic design. Once the safety of the robot is confirmed, it goes into the manufacturing and programming of the robot system. After the design drawing of the entire robot system is completed, it goes through manufacturing, testing, delivery, and installation, and then proceeds to the phase of production operation. But even with a successful deployment, the job of a robot manufacturer or system integrator isn’t over. The company has a long relationship with the end-users that use the system, including regular inspections, customer support, and assistance if failures occur. Kawasaki Heavy Industries has a dedicated call center to answer any questions end users have after installation. There is also a 24-hour help desk for problems that arise outside of business hours. Another reason for Kawasaki’s popularity among users is its extensive follow-up and customer service. Kawasaki Heavy Industries’ after-sales service team was established more than 30 years ago. And in 1986, it established Kawasaki Robot Service Co., Ltd. (formerly Kawasaki Robotics, Ltd.), a company specializing in maintenance and after-sales service. Behind the Kawasaki robots, there is always a team of experts close to the robot’s life, from installation to operation, maintenance, and renewal.
Konceptuelt billede af en robot, der erstatter mennesker
There are many reasons why customers choose Kawasaki as a partner in the introduction of robot systems. One of the reasons for this is that, as a company that started as a manufacturer and has deep roots in this industry, it is fully equipped to support the introduction of robots to customers. For example, the Nishi-Kobe Plant has one of the largest robot showrooms in Japan. There are vertically articulated robots, parallel link robots, clean robots, and even duAro cobots and Successor systems. In the showroom, you will see work environments replicated for each robot type, such as welding, painting, and sorting lunch boxes on a production line. Many users want to implement robots to help combat labor shortages. Or, they may want to increase production efficiency and diversify their product line. Another common reason is to prevent human error and improve product quality, or protect workers from harsh and dangerous work. Every company has its own incentive to think about robots. Industrial robots are certainly the best solution for these problems, but replacing humans and robots is not enough. There’s no such thing as a robot without a professional standing by and supporting from the beginning to the end. Automation systems can only run smoothly when they have both robot system integrators and robot manufacturers.

[Columns]
Introducing Industrial Robots Faster!
Start of K-AddOn operation

Industrial robots cannot work by themselves. They need to connect to peripheral equipment such as grippers and vision systems so the whole system can work. In order to smoothly connect devices made by the various manufacturers, it is necessary to connect and link the respective software types. The K-AddOn platform was launched by Kawasaki to speed up the time it takes to connect the robot to its peripheral equipment and help ensure a smooth deployment. By opening the interface of industrial and collaborative robots made by Kawasaki Heavy Industries to peripheral equipment manufacturers, the robot system integrator and end-user can reduce the verification cost of equipment connection required at the time of installation.
Et diagram over K-AddOn, en platform lanceret af Kawasaki Heavy Industries for at forkorte forbindelsestiden mellem robotter og eksterne enheder og understøtte en problemfri implementering.

K-AddOn

Ready to take the next step toward automation?

]]>