Armar 3

from Wikipedia, the free encyclopedia
Armar IIIb

Armar 3 is a fully autonomous and adaptive humanoid household robot developed by the Karlsruhe Institute of Technology , which can find its way around in any kitchen and is supposed to do everyday tasks there.

Project data

General

Furnishing

  • High-performance computer: 5
  • Controls: 12
  • Motors (for the drive): 2 × ≈ 95  watts
  • Cameras: 4 color cameras (2 per eye)
  • Microphones: 6
  • Area laser scanners: 3 3D area laser scanners

costs

  • Material costs: 200,000 euros
  • Funding: 8,000,000 euros

Details

Armar 3 was specially designed for use in the kitchen to relieve people of annoying everyday tasks (such as loading the dishwasher). Some functions, such as loading and unloading the dishwasher, are only available to him and his brother.

Drive and energy supply

In contrast to many other humanoid robots, Armar 3 does not move bipedally , but by means of three omnidirectional wheels that are mounted on the underside of the chassis offset by 120 °. They enable Armar to move easily in all directions. The drive is provided by two 24 volt collector direct current motors with an output of around 95 watts each. There is also a reduction of 1:36 by means of an upstream planetary gear . The necessary energy - Armar moves completely autonomously, without any power cables - Armar draws from three lead-acid batteries . They are also located in the chassis. In addition to electric motors, Armar also uses other working bodies. The robot hand contains fluid actuators (movable chambers that are filled with a gas or a liquid and then deform), which have the main advantage that they are small and light, but the hand is also flexible and therefore less risk of injury exists for humans. But Armar also makes it possible to keep raw eggs without any problems.

Armars finding his way around the kitchen

As mentioned above , Armar has three 3D area laser scanners , with the help of which he creates a 3-dimensional image of his kitchen, which even includes people. The three scanners, which were manufactured by the company Sick AG , are each offset by 120 ° between the three rollers at a height of less than 15 centimeters in the robot's chassis so that they can even perceive the tips of a person in the room. Thanks to the scanning angle of each scanner, which is 180 °, Armar 3 is able to “see” its entire environment. The range of the laser scanner is around 50 meters, the resolution is around 0.5 arc seconds. So that Armar's 3D model of the kitchen is always up to date, the 3D surface laser scanners produce 30 scans per second. For the acquisition of further position data, Armar 3 is equipped with optical incremental encoders , which record the angular position and speed of the wheels with a resolution of 500 pulses per revolution.

Communication with Armar 3

Armar 3 responds to both English voice commands and pointing gestures. In addition, he can speak himself, also in English. He perceives the acoustic signals using six microphones. In doing so, it recognizes other disruptive sound sources as well as important sound sources that arise in a kitchen. He can calculate these out and thus clearly perceive the voice commands. Armar 3 can also precisely locate in three dimensions who is speaking and where he / she is speaking. He perceives the pointing gestures using his four color cameras - one for close objects and one for more distant objects per eye - with the help of which he can also see three-dimensionally. The eyes are also used for face recognition or the camera tracking of faces and can even locate people, but are not used for navigation in the kitchen. The area laser scanners are solely responsible for this (see Armar's way of getting around the kitchen ).

"Knowledge" of Armars

Since Armar should react to speech signals and then also have to know what to do, Armar 3 has a kind of database in which all objects - cups, mugs, packages, and so on - are stored as many images from several, different perspectives so that Armar can find them in his kitchen. For each object recorded in the database, however, it is not only stored what the object is called and what it looks like, but also how it can be treated, what properties it has and what uses it is. For this purpose, certain movements of an object were recorded using a data glove and motion capture . To do this, you grasp a cup with your hand or the data glove and, among other things, perform maximum movements with this object, i.e. you hold the cup at the maximum angle of inclination, or move it with maximum acceleration in a certain direction. Based on this data, Armar 3 always knows how to handle the gripped object. But the individual movements - stretching out your arm, grasping objects, ... - are stored in the robot. But not everything had to be written into its program at the beginning, but Armar 3 is able to imitate and save movement sequences that are presented to it by a person, or to then also repeat them in the corresponding situations. When recognizing the movements of the person who is showing them to him, he or she is guided by the contours of his counterpart. Another possibility for the scientists is to use a data glove and motion capture to record individual movement sequences, including the tablecloth, in three dimensions and finally to feed the robot with the data obtained. This previous demonstration of the movements by humans also causes the robot to move more human-like afterwards.

Armar's skills

Armar 3 is also able to bring you certain smaller items - for example cups, mugs, rice or juice packs. Fetching a certain drink from the refrigerator is no longer a big problem for Armar 3. He recognizes which things he should bring from the packaging sample. But he cannot identify new objects that he does not yet know. Setting the table is also one of its functions. Among other things, he even masters loading and unloading a dishwasher. No other robot can do this.

goal of the project

According to DFG ScienceTV, the aim of the project is for Armar 3 "to [...] at the end [...] independently recognize what needs to be done in the household and do it".

Web links

Commons : Armar  - collection of images, videos and audio files

Individual evidence

  1. ↑ a b c d e f Design and construction of a holonomic, mobile drive platform for a humanoid service robot ( Memento from July 5, 2010 in the Internet Archive ), PDF file about the chassis of Armar 3, ODETE and other predecessors (Armar, Armar 2 )
  2. ↑ T. Asfour, DN Ly, K. Regenstein and R. Dillmann: Coordinated Task Execution for Humanoid Robots http://link.springer.com/chapter/10.1007%2F11552246_25 , 2006
  3. ↑ a b c Helpers in the household - W for knowledge. In: daserste.de. Retrieved July 18, 2019 .
  4. ↑ a b c d e f g h DFG ScienceTV - The Human Machine ( Memento from June 2, 2009 in the Internet Archive ), content from the 12 videos (individual listing of which information is in which video, in the oldest version of this article)
  5. ↑ a b c d e Video turntable Germany - digital household help  in the ZDFmediathek , accessed on April 22, 2008. (offline)
  6. ↑ DFG ScienceTV - The Human Machine ( Memento from June 2, 2009 in the Internet Archive ), Video 01/12