It's Dangerous When You Build Things You Klutz, but MIT's Robots Will Save You
Remember when all furniture was custom made? Yeah, me neither. These days, it's all about affordability through mass production. I’m as much of a fan of modernist Scandinavian design and cheap meatballs as anyone, but mass produced furniture is by definition a compromise: It's sort of okay for everyone, which means it's never exactly what you want. But custom made furniture that's perfect for your space and style is expensive, because it takes a substantial amount of skill and equipment to produce.
At MIT's Computer Science and Artificial Intelligence Laboratory, which is presumably populated by researchers trying to afford furnishings for their overpriced and undersized Cambridge apartments, researchers have developed a system called AutoSaw that (nearly) solves the entire furniture problem. AutoSaw can help you pick out a piece of new furniture, customize it to the shape and size that you want, and then build it with a team of robots to your exact specifications.
Users design their furniture using an interface (top row) that also displays the stress distribution on the pieces (middle) and their elastic deformations (bottom).Image: MIT CSAIL
AutoSaw consists of four stages. The first involves coming up with furniture templates using a special design application. The researchers say this step is best left to human experts who actually know a thing or two about carpentry, since it involves coming up with designs that are feasible to be manufactured and assembled with materials and techniques available.
Once you have the templates, you get to go in and mess with them a little bit. AutoSaw includes a user-friendly interface that provides sliders to adjust parameters like leg length and spacing, and the size and shape of the top surface for tables. You can also futz with parameters for templates for chairs and sheds, and if you’re very ambitious, even an outdoor deck. When you have the design the way you want it, AutoSaw simulates stress distribution and elastic deformation for you before everything gets finalized.
Step three is where the robots come in. With power tools. Now you, as the accident-prone consumer, only need to stand back!
Here's how the researchers, a team from MIT's Distributed Robotics Lab led by Professor Daniela Rus, explain in a forthcoming paper how AutoSaw works:
We used two robotic systems to automate the chop saw and jigsaw processes. The jigsaw process was automated using a previously developed robotic jigsaw. This jigsaw robot is a modified Roomba Create with a jigsaw installed in the center. It uses a Vicon positioning system for state estimation and a previously developed MPC and planning algorithm to perform the cuts.
The chop saw process requires multiple robots to automate. Two Kuka Youbots lift lumber and place it on the chop saw. Each Youbot was outfitted with special complaint grippers. The grippers allow the robots to clamp onto material, to drive the material along a direction, and are complaint perpendicular to the major axis of the lumber. When lumber is placed in the gripper, a force perpendicular to the lumber and parallel to the gripper can cause the gripper to shift its grip. The chop saw is automated by attaching a relay to the 120V line and a linear actuator is attached to the saw.
The output of AutoSaw at this point is a pile of parts that still need a little bit of work before they can be assembled. The system doesn't include drills, so the peg-and-hole connections involved a manual step that the researchers hope to eventually automate. We should also mention a few other constraints: The jigsaw robot is only able to cut through foams (although it's getting an upgrade to be able to handle plywood), and the youBots aren't strong enough to lift pieces of lumber much bigger than 1x2 or 1x3.
AutoSaw generates assembly instructions automatically (including step-by-step visual diagrams), based on how you defined the connection points in the original template. The system doesn't berate you during the assembly process, making it somewhat unrealistic, but you can usually rely on friends or family for that anyway.
For some additional detail, co-lead author Jeffrey Lipton, a post-doc at MIT CSAIL, answered a few of our questions via email.
IEEE Spectrum: Why does an expert need to be a part of this process? Can't the design of furniture templates be automated somehow?
Jeffrey Lipton: Simply saying "make me a table" to software is something that is too hard to do right now. The context is ill-defined. Are we making a coffee table, or a dining room table, or a end table? What style is it in? What are your personal likes and dislikes? Where will it be? What's around it? We use expert designers to make templates in order to parse the space of all possible table designs into something a layperson can explore more easily.
In what situations would the simulation step provide me with information that would cause me to go back and alter my design parameters?
Take the deck example. Looking at the simulation could tell you you must construct additional pylons—you simply wouldn't be able to support enough people on the deck without them.
Why did you decide to use a stationary chop saw and two mobile robots as opposed to a mobile robotic saw?
Right now there are industrial machines that support a saw at the end of the robot arm. However, supporting such a large mass through such a long kinematic chain makes the robots heavy and expensive. Instead we took our inspiration from human carpenters. Humans don't hold out a circular saw at arms length to cut. Rather than use long kinematic chains, they use local reference points on the power tool to achieve accurate cuts. By using the same technique, we can use much smaller and lighter mobile robots.
We saw IkeaBot assembling furniture in 2013. At this point, what would it take to go from raw materials to assembled furniture without human intervention?
End-to-end production without humans isn't necessarily our goal. Right now we are focusing on keeping the human hand away from potentially dangerous items like saw blades. There is still a long way to go using the mobile robot platforms. We need to add tools, especially the drill (to make the parts capable of assembling). We will also need to make the robots able to install some of the hardware like threaded inserts. We still haven't integrated sanding and painting into the system as well.
You say that you’re working towards "enabling mass customization of carpentered items." What's your vision for the future?
We imagine that in the future consumers will no longer have to rely on a fixed number of different kinds of furniture for their homes. If you live in a cramped apartment in New York City, you won't have to worry about whether conventionally-shaped furniture will fit in your place— you’ll be able to customize the exact size and shape of any chair or sofa or table based on what works for you.
IEEE Spectrum: Why does an expert need to be a part of this process? Can't the design of furniture templates be automated somehow? Jeffrey Lipton In what situations would the simulation step provide me with information that would cause me to go back and alter my design parameters? Why did you decide to use a stationary chop saw and two mobile robots as opposed to a mobile robotic saw? We saw IkeaBot assembling furniture in 2013. At this point, what would it take to go from raw materials to assembled furniture without human intervention? You say that you’re working towards "enabling mass customization of carpentered items." What's your vision for the future?