Choreographing the movement of robots in automated manufacturing processes is complex, time-consuming and expensive. This new algorithm may be a solution.
Robots have become an important part of the automotive industry – think of an auto plant and you’re probably picturing articulated arms moving in complicated patterns. However, the tireless efficiency and precision they offer comes at significant cost – a tremendous number of man hours are required each time a new assembly line is set up or an existing one adjusted. The arms themselves have complex geometries, with multiple joints and end-of-arm tooling which need to be calibrated to sub-millimeter precision. Mistakes can mean the difference between a complicated dance of productivity and expensive chaos.
This fact puts a hard limit on what can practically be automated. Every movement needs to be carefully programmed – not only so that the robot can do its job but also to avoid having one robot smash into another. The challenge of choreographing more than one or two robots in a shared work cell can quickly stretch beyond what human technicians are capable of. The problem simply grows too complex to tackle manually.
As with most manufacturing and automation challenges, engineers all over the world have been trying to find the most efficient way through this bottleneck, and researchers at Boston-based automation programming start-up Realtime Robotics believe they’ve found a solution.
“We’ve solved motion planning” says George Konidaris. Along with his co-founders and team at Realtime Robotics, Konidaris is working to “broaden the envelope of what robots are able to do in real life, which is actually very narrow at the moment.” The result? Robots that are “more adaptable, flexible and easy to program.” In some ways, a little more like us.
A professor of computer science at Brown University, some years back Konidaris shared a “fortunate lunch” with a like-minded colleague and future Realtime Robotics co-founder Dan Sorin, a computer architect. He shared Konidaris’ interest in motion planning and creating a more intelligent robotics system. They also brought the right skillsets together to tackle this problem. It would take groundbreaking work with uniquely designed algorithms and processors to make these sorts of calculations fast enough for practical use. Two of their students, Will Floyd Jones and Sean Murray, were also there from the start and served as additional co-founders.
The result was an approach that is miles away from the current state of the art. What used to take days to program manually can now be accomplished in mere hours. In turn, it becomes possible to have more working arms operating in the same cell, saving cycle times and costs.
“With our solution you can just input what you want the business ends of the robotic arms to be doing. We can deconflict all six robots and keep them moving at maximum speed. They won’t crash into each other. This is simply beyond the ability of humans to do manually.”
Konidaris offers a useful analogy: “You don’t tell your home printer how fast to move the head and where it should be at one time. You simply tell it what you want to print. That’s what Realtime Robotics can make possible on the factory floor.” In other words, they’ve added an analog of the sort of general physical intelligence you and I rely on each day. “If you want to reach into your fridge to grab a can of beer, you don’t spent days thinking really hard about each movement you’ll make” he notes.
This isn’t to say that people immediately recognized the importance of this work. Indeed, many in the industry consider the problem of robotic movement to be solved. Konidaris acknowledges that this is in large part because of very advanced work that has already taken place using conventional means. “There are a lot of very skilled engineers who have built up a well-established practice around this,” he notes, saying it’s a testament to the utility of robots on the factory floor, even with all the challenges that have historically come along with them.
The key was finding the right audience to get their foot in the door. Toyota AI Ventures would prove to be the group that first saw the potential, immediately understanding the strategic importance of this new capability. Konidaris credits Toyota’s emphasis on motion planning research – the company already understood there was significant room for improvement with how robots operate on the factory floor. Since then, other corporate partners have followed. Most of them need to remain confidential, but BMW was one of their other early big catches.
The significance of the work really can’t be overstated. The robot motion planning technology exploits parallelism and a carefully co-designed algorithm. It allows engineers on the shop floor to work at a higher level of abstraction. Rather than worrying about the movement of each joint in a six or seven-jointed robot arm, they simply need to know where they want the working end of the arm to be at a given time. “It’s like an operating system. It handles all the deconflicting problems and makes sure there are no mashups in the space,” Konidaris says. The upshot is that you could have as many as 12 arms operating in one work cell. Like the printer or beer can analogy, you just need to tell the system what you want to do, and it will handle the complex geometry.
The team is hopeful that this technology will one day dominate the automotive industry. Konidaris is particularly optimistic, suggesting that in the next five years most of the automotive industry will be using the technology. He sees this as a means to push the cost of manufacturing down, as more of the assembly process can one day be automated.
After that, other industries are next. Everything from pallet unloading to semiconductor production could use this type of technology to increase efficiency. Indeed, the Realtime Robots team is already looking in these directions. Konidaris even hints that this is a step towards a more general intelligence in robotics. One day, you might not even need to rely on working components always being in the exact same space. Picture a robotic arm that can sort widgets by color or adjust on the fly to changes in its environment, such as misplaced materials or a human walking by.
