- Success Stories
- 3D Scanners
- About Us
3D Scanning Technology Helps Robot Grasp Reality
What could you do with a can of chopped tomatoes, marjoram spice, ravioli, and sauerkraut? You could ask your humanoid robot to make you dinner! Today, robots need 15 hours to make one dish of pasta, but European university researchers at the Karlsruhe Institute of Technology (KIT) in Germany are working to change that. They are developing ARMAR-III, a robot that learns and understands its surroundings by recognizing and grasping 3D modeled-objects within a controlled lab kitchen environment. Currently, ARMAR-III is comfortable opening and closing a dishwasher or even making a simple meal, however the long term goal is to create a personalized system of humanoid robots working alongside humans to create a more productive and higher quality future.
Thousands of homes already have Roomba robots vacuuming their floors. Creating robots that interact with, and assist humans in a reliable and safe manner is another thing entirely. The KIT research team’s goal of cooperating and entering a dialogue with ARMAR-III vastly transcends your typical android. This robot needs to understand both what it perceives and what it does. “If you want to interact, the robot needs data – what an object looks like and how to pick it up,” explains Alexander Kasper, Dipl. Inform., Humanoids and Intelligence Systems Lab – Institute for Anthropomatics at KIT. “Our main focus is to communicate with the machine through speech and gestures, eventually increasing the robot’s cognitive abilities to learn new things and survive in a human environment that is changing all the time.” He is referring to some of the toughest challenges in modern robotics, namely visual object recognition and machine learning. These are abilities that Kasper says we humans develop naturally, even if we need a few years of training. The key challenge in teaching ARMAR-III how to survive is by learning through doing.
How does a robot “learn”?
The ARMAR project has been developed during the course of the Collaborative Research Center 588 (Humanoid Robots) – Sonderforschungsbereich 588 (SFB-588), led by Prof. Dr.-Ing. Rüdiger Dillmann. Prof. Dillman’s team teaches ARMAR-III to understand its surroundings by recognizing and grasping objects, which requires the robot to have the right 3D modeled data. ARMAR-III can find and pick up a box of mashed potatoes amidst many other similarly shaped objects because the robot has learned how to interact with this particular box. The weight, colors and contents of each object are unique to all of the 119 objects that Kasper has modeled in ARMAR-III’s lab kitchen environment.
Before the KIT research team discovered Rapidform.dll, they were trying to use open source software and the software that came with the team’s 3D scanner. These options were too basic and did not integrate well into one good workflow, which created more cumbersome 3D modeling work.
At the beginning of 2008, Kasper’s colleague was looking for 3D scan data processing software for a separate project, and found Rapidform XOR. Impressed with its capabilities, he recommended that Kasper look into using it for ARMAR-III. Kasper was thrilled when he learned that much of Rapidform’s functionality is available via Rapidform.dll, a Software Development Kit (SDK). The Rapidform.dll SDK makes much of Rapidform’s world-class point cloud and mesh processing tools available to anyone to add to their own 3D applications.
The KIT team created an object modeling center with a Konica Minolta 3D scanner, a turntable and a pair of moveable Marlin cameras to fully scan real-world objects and combine the output into one textured 3D image. The modeling center was funded partly by the SFB-588 and project DESIRE and the results of the modeling center have been used in both projects as well as some European research projects.
It has been Kasper’s responsibility to model the environment in which ARMAR-III learns and lives. He has modeled 119 objects available to ARMAR-III with “good resolution”, all saved online in the KIT Object Models Web Database for public use. These high quality 3D modeled objects allow for grasp planning – ARMAR-III computes grasping points and what is a safe grip entirely independently, no human input is involved. Kasper said the KIT custom-made software “greatly reduced work we need to do ourselves, I don’t think we would have the level of quality without Rapidform.dll,” and it “worked great, I’m very happy with it.”
“We are committed to arming 3D application developers with powerful tools that can be seamlessly integrated into their existing software”, says Michelle Baek, Rapidform.dll Business Development Manager, “to reduce their development time so they can focus on providing world-class 3D scanning solutions.”
The ARMAR “family” of humanoid robots has been in development since 2000. Each generation received a boost in degrees-of-freedom (DOF), mobility, sensory capabilities and copycats among other European-designed robots. Yet the trailblazer ARMAR-III currently still resides on an autonomously driven mobile wheel platform. “ARMAR-IV will feature legs and learn to walk,” says Kasper. “The robot is currently limited with dexterity, but with legs it will be able to climb steps and provide more assistance to humans, similar to Japanese robotics technology as seen in the Honda Asimo.” Does he ever worry about Artificial Intelligence evolving to the point of a robot apocalypse? “In 80+ years there will be a personalized system with robots in each home….or I, Robot” Kasper laughs. “Robots will eventually be able to use their learned knowledge to guess what the user wants. Even today, a human does not operate ARMAR-III; instead the robot reacts to the human.”
The Karlsruhe Institute of Technology would like to thank the Deutsche Forschungsgemeinschaft (DFG) as part of the Collaborative Research Center Sonderforschungsbereich 588 (SFB-588) project for funding ARMAR. Their object modeling center was funded partly by the SFB-588 and German Service Robotics Initiative “DESIRE”.