A research team from the Massachusetts Institute of Technology (MIT) was nominated for two best-paper awards at the International Conference on Robotics and Automation this week.
The annual event, hosted by the Institute for Electrical and Electronics Engineers, recognizes achievements made by scientists in the field of robotic engineering.
The MIT study features a new algorithm that is capable of significantly reducing the planning time of robot teams. While the resulting plan made by the algorithm may not be perfectly efficient, the planning time the process saves is more than enough to offset the execution time it adds.
"We're really excited about the idea of using robots in more extensive ways in manufacturing," MIT professor Daniela Rus of the electrical engineering and computer science department and co-developer of the new algorithm said.
"For this, we need robots that can figure things out for themselves more than current robots do. We see this algorithm as a step in that direction."
In order to test the feasibility of their algorithm, the researchers used it to guide a crew of three robots to assemble a chair.
One of the issues the MIT team was able to address involved teaching the robots how to follow a series of discrete steps needed in an assembly operation. This often requires collaboration between groups of robots.
When the test began, none of the robots knew which particular step it would be assigned to. The entire process was planned in real time.
The challenged lied in distributing specific tasks to each robot during assembly. They should also avoid colliding with each other when they reach the collaborative phase.
What normally took up much of the planning time was finding out a way for each robot to efficiently grasp the object it was supposed to manipulate in to complete its own task as well as the other ones that would follow it.
Rus explained that the configuration for the grasp action could work on the current step but it could also prove to be problematic for the next because it would often need a different sensor or robot. She said the current grasping formation might not allow another robot or sensor to join in on the task.
To solve this, the team made use of a multiple-step assembly operation and optimized the placements of the robots that allowed each one to finish the entire process.
What makes the new algorithm different from others is that it can defer most difficult decisions associated with grasp position until it finishes all the easier ones. This allows the process to be interrupted at any part, but it will still end up with a viable assembly plan.
The robots may at times have to drop the objects they were holding and pick them up again, especially if the algorithm did not have time to calculate an optimal solution. However, the time it would take for the robots to do these actions would be trivial compared to the time the algorithm would save in producing a comprehensive solution.
The algorithm began solving the issue with the assembly operation by disregarding the grasping problem completely. It then calculated every step needed for the operation from the point of view of a single robot and a single part of the chair being assembled.
The program went through each step, determining whether a particular part could be used to finish two stages having to modify the behavior of the robot. It chose which ones would work within the framework of the assembly process and postponed those that would not.
After finishing the easy parts, the algorithm then revisited the ones it had postponed. It modified the behavior of robots to effectively smoothen the transition between stages.
If allowed to reach the completion of its task, the algorithm would end up revising the behavior of all the robots at every phase of the assembly, resulting in an immensely complex task. The best process would be to let the robots drop what they were holding at certain phases instead of allowing the program to calculate an optimal solution.
Bradley Nelson, a professor of robotics and intelligent systems at Zurich's Swiss Federal Institute of Technology, said that the MIT algorithm was able to show how robots can be guided to solve a complex planning problem such as assembling a chair.
Nelson said the only problem with it is that it will not allow him to assemble his Ikea chair himself at home.
