Developments in autonomous robotics have the potential to revolutionize manufacturing processes, making them more flexible, customizable, and efficient. But coordinating fleets of autonomous, mobile robots in a shared space – and helping them work with each other and with human partners – is an extremely complicated task.
Researchers at Stanford have created an algorithm that can take a design plan for a particular product and figure out the most efficient way to manufacture it with a team of robots. Their work, published recently in the journal Robotics and Autonomous Systems, includes planning how to construct subassemblies that are built separately and then combined, such as constructing a car door and then attaching it to the body; directing the robots to work both alone and in teams; and laying out the assembly floor in an efficient manner that prevents collisions.
"What's really unusual about what we're doing here is the scope of the problems we're solving," said Mac Schwager , an associate professor of aeronautics and astronautics at Stanford and co-author of the paper. "There has been research into some of these individual pieces, but I think we're the first to really think about how it all fits together into a large-scale system."
Modular manufacturing
The ability to generate assembly plans quickly and efficiently could help provide a new level of flexibility in manufacturing. Currently, automated assembly lines are very rigid – they can build one thing quickly and well. Using general purpose robots and distributed stations that are able to accomplish basic manufacturing tasks, such as welding or sanding, factories could be able to pivot more quickly or create customized products without having to retool the entire manufacturing floor.
"Right now, if you want to change your construction pipeline to something different, it requires a lot of planning and work to tear it down and set it back up," said Dylan Asmar, a PhD student in the Stanford Intelligent Systems Laboratory and co-author on the paper. "With a more modular approach like this, changing your pipeline would be a lot easier and more streamlined."
To make this modular construction process a reality, manufacturers need to be able to rapidly plan, coordinate, and reconfigure the movements of robots around the factory floor. Asmar, Schwager, and their colleagues designed an algorithm that can do just that. The researchers tell the algorithm how many robots it has to work with and the basic specifications of those robots, such as how much each one can carry, and provide a schematic of what they want to build and the manufacturing tasks that need to occur. The algorithm determines how the robots will split up to construct subassemblies that can be built separately from each other and how the robots will bring these pieces together quickly and efficiently.
"Our objective is to go from raw material to the finished product as quickly as possible, and the way you do that is through parallelization," said Mykel Kochenderfer , an associate professor of aeronautics and astronautics at Stanford and senior author on the paper. "It's not a linear sequence – we try to do operations in parallel as frequently as possible."
The algorithm lays out assembly stations and assigns specific robots to collect and deliver parts to the correct stations at the correct times. It directs the robots to work in teams when parts are too large for an individual robot to carry and maps out how the robots will move to avoid interfering with others. And it does this all remarkably quickly – it took less than three minutes for the researchers to generate plans to assemble a toy construction block model of a Saturn V launch vehicle, which has 1,845 parts and can be broken into 306 subassemblies, with a team of 250 robots.
A platform for experimentation
"There are still plenty of problems to be solved before our work could be used in a real-world manufacturing context," said Kyle Brown, who began this work as part of his doctoral thesis and is the lead author on the paper. Brown and his colleagues have built a simulator to help other researchers test their own construction algorithms and bring the next revolution in manufacturing closer to fruition.
The open-source platform allows researchers to try out new algorithms or adjust existing ones to see how optimizing certain aspects or working within specific constraints affects the process as a whole. It evaluates those algorithms with toy construction block models. Brown has also used the simulator as an educational tool for elementary school students, letting them race against the robots to construct a model of an airplane.
"I adjusted the speed of the simulation so that the robots went slow enough for the kids to just barely win," Brown said. "The kids were elated at their narrow victory, and I got to teach them a little bit about robots. They may not all grow up to be roboticists, but this was definitely a positive exposure to the field."
