Nov.19, 2014 | By Kira
A team of researchers at Oslo University's Department of Informatics has developed a category of robots that, through simulation programs and 3D printing, adapt to unforeseen obstacles. In turn, these self-improving robots can enter hazardous areas and perform tasks in situations too extreme for human involvement, such as radioactive sites, deep mines, or even distant planets.
Image: University of Oslo; Apollon Research Magazine
The real-world uses for these robots are vast. One possible application, as described ay Associate Professor Kyrre Glette, imagines a robot entering a wrecked power plant and stumbling upon an unforeseen staircase. The robot is able to take a photograph, analyze the situation, and, using its integrated 3D printer, produce a new part, or even a new robot, that can negotiate the stairs. The entire scenario plays out without human interaction and without endangering or delaying the mission.
Oslo University's Robotics and Intelligent Systems research team has been investigating these possibilities for years, and has already developed three generations of self-learning robots. The first robot, affectionately named "Henriette," taught itself how to walk and jump like a chicken. When it lost a leg, it quickly adapted and learned how to hop on the other.
The second generation, developed by Master student Tønnes Nygaard, was almost completely self-designing, using a simulation program that could calculate how the robot's body should be shaped, the length of its legs and the optimal distance between them.
The third and most flexible generation to date takes the simulation process one step further. Researchers can input their desired characteristics (such as the speed, size, energy consumption and capabilities of the robot) and then utilize the simulation program to run through thousands of possible configurations. The program suggests the optimal number of legs and joints, taking in a number of considerations.
Finally, the best possible models are selected through a process of artificial evolution. Ultimately, researchers are left with a set of proposals for different robots designed for optimal functionality, which can then be 3D printed and tested by real-world standards.
Image: University of Oslo/Robotics and Intelligent Systems
The 3D printers used by the University of Oslo cost between NOK 400,00 to 3,000,000 (roughly €47,000-€350,000), with the more expensive models capable of combining milling and layering techniques to ensure the highest level of precision.
Oslo University researchers have praised 3D printing as an invaluable technology that allows them to construct seemingly impossibly complicated structures as a single piece. Not only is 3D printing essential to creating robotic parts, it can also be built into the robots themselves as a self-enhancing tool, as in the staircase scenario described above.
In addition, 3D printing allows for a reduced timespan between the ideas stage to the robot-testing stage. Improvements can be made efficiently and with as little delay as possible.
However, the transition from simulated design to 3D printed robot is not perfect. The chief obstacle faced by the Informatics team thus far has been closing the so called "reality gap" between what the simulated robots can accomplish, versus their performance in the laboratory.
"Once the robots have been printed, their real-world functionalities quite often prove to be different from those of the simulated versions," explains researcher Mats Høvin, "We are therefore studying how the robots deteriorate from simulation to laboratory stage."
Image of Mats Høvin, member of the University of Oslo Robotics team Image: University of Oslo/Robotics
In order to close this gap and to create robots that are as good as, or even better than their simulated counterparts, researchers have set up obstacle courses to test how their robots can self-teach and self-adapt when confronted with new obstacles.
According to the researchers, real-world practice enables the robots' performance to be improved by 20 to 40 percent.
Ultimately, researchers hope that future models will be completely autonomous and, in fact, evolutionary. Much in the same way that our bodies send information to our brains, future robots will provide automatic feedback to the simulation program, enabling computers to create increasingly improved versions.
Despite the achievements made by the Robotics team in Oslo, further work is needed before the robots can be exploited commercially. "Our greatest challenge is to develop robust algorithms and a system which is able to make use of imprecise simulations," said Associate Professor Glette.
Nevertheless, given the possibilities and high turnover rate provided by 3D printing, it may not be long before these autonomous robots are sent where no man has gone before.
Source: University of Oslo
Posted in 3D Printing Applications
Maybe you also like:
- Pero: beautiful 3D printed pigeon bookmarks
- 3D printer 'gun parts' found in Manchester raid may be parts for a printer
- UK police find first '3D gun-printing factory'
- Impressively complex 3D printed Vulpes Exploration Rover Alpha 1
- First 3D printers arrive to Haiti
- Disney sees 3D printer in every home within a decade
- Hollywood uses 3D printing to create monsters nowadays
- 3D printed "Smoke Dress" hits the Volkswagen catwalk in Frankfurt
- 3D printing to dominate STEM education within 3 years
- 3D printed Sunshade makes Google Glass easier to see in the sunlight