Test Bed Builds Up Robotics Research at Carnegie Mellon University

The headquarters of Carnegie Mellon University’s Manufacturing Futures Institute (MFI) is like a playground for manufacturers. Its facility at Mill 19, a former steel mill just outside of Pittsburgh, Pennsylvania, features high and low bays with CAD/CAM programming, a five-axis CNC machine, wire EDM, laser cutting, an additive manufacturing lab, metrology equipment, welding equipment, robotic arms, autonomous mobile robots (AMRs) and…bins of multicolored Lego bricks.

The Lego bricks are part of the MFI’s robotic test bed, where researchers can test and study cutting-edge robotic applications. AMRs can be used to deliver bricks to the four small Yaskawa robotic arms mounted to tables, which then perform tasks assigned by researchers. The robotic arms are also used to study other applications, such as insertion of connectors on robotics task boards designed by the National Institute of Standards and Technology (NIST). The whole area is monitored by sensors and light curtains for safety, so humans can share the space with the robots.

I recently had the chance to visit the MFI facility and speak with researchers there. While today we typically associate robots in manufacturing with repetitive tasks such as machine tending, assembly and welding, CMU researchers’ work offers a glimpse at tasks robots will play in manufacturing facilities in the future.

Learning Through Play

Why use Lego for the tasks in the test bed? The answer is both simple and complicated. Lego is a simple, yet flexible concept. They can be assembled into many different structures, disassembled, and re-assembled into something new. Plus, many people are familiar with them. “I think Lego makes it very easy to show the concept to non-experts because almost everybody has played with Lego before,” explains Changliu Liu, an assistant professor at CMU’s Robotics Institute and director of its Intelligent Control Lab. At the same time, the bricks can be challenging to manipulate. The robot needs to ensure the bricks are fully connected when assembling a part, and carefully remove bricks one at a time during disassembly without destroying the rest of the structure. CMU researchers even designed a special 3D printed gripper based on a brick separator tool for the robots to use. Plus, the way Lego bricks are assembled isn’t dissimilar to tasks required for assembling printed circuit boards.

The whole point of the test bed is to provide a flexible environment that can handle many different tasks. “The idea is to have a simulated physical space for a manufacturing shop,” explains Shobhit Aggarwal, an advanced manufacturing engineer at MFI who oversees the test bed. “So, it’s targeted towards high-mix, low-volume manufacturing where you would have different kinds of assemblies and not just one continuous production line.”

The test bed itself serves multiple purposes. Not only can it enable researchers to test robotic applications, but it can also generate data without human intervention. “The idea is to provide a research base for doing digital twin and manufacturing-related research,” Aggarwal says. “All of the individual resources included in the test bed have a digital twin and simulation aspect.”

Digital twins help not only with monitoring, but also with planning, testing, and ensuring that everything is safe and the equipment is protected. “The majority of the use of digital models has been to test out algorithms, test out policies and test out planning trajectories,” Aggarwal notes. They can also help during live visualizations, giving enhanced information beyond what a camera can provide. This will help in addressing another challenge: figuring out how to capture humans within the environment. He says the next step is to use depth cameras to capture humans in environments with robots and create a mixed-reality environment.

Research Projects

The test bed was completed in late summer 2023. Since then, it has been used in a range of research projects:

• Generative AI and Robotics. “We developed a system where, using generative AI, human users can say a description of what assembly they want, then the generative AI can turn that into a digital file,” Liu says. Then, the robotic test bed builds the assembly. So far, she says it has been successful with small chair and pavilion-type structures.
• Multirobot Collaboration. Other research has involved two robots working together in the same environment or on the same task. Frequently, Liu says, one robot is doing the main action (plugging something in) while the other robot stabilizes the workpiece. But if the application calls for two robots to be doing the same task separately, or two different tasks, they need a “conflict resolution system.” Philip Huang is a second-year Ph.D. student studying robotics at CMU whose research focuses on multirobot collaboration. He says to envision a huge warehouse with many robots moving around at once, all trying to fulfill orders. Or a manufacturer with multiple robot arms in one space trying to complete assembly tasks. His research focuses on not only ensuring that robots don’t collide, but also how to best allocate resources to complete tasks in the most efficient way.
• Vibrotactile Sensors for Assembly. Kevin Zhang is a fifth-year Ph.D. student in robotics at CMU whose research focuses on using vibrotactile sensors in assembly tasks. Vibrotactile sensors are a type of microphone that can record sound data. This data can be used to confirm that assembly tasks have been completed correctly. When a USB connector is plugged in or a screw is fully screwed into a part, there’s a distinct sound. “Our goal is to hear those contact changes and then predict the successes so we can move down and continue the manufacturing process,” he explains. These sensors have the potential to provide more accurate data than vision sensors.
• Light Curtains for Safety. Traditional light curtains have a fixed infrared emitter and receiver. When the light curtain is broken, the receiver stops receiving the infrared signal and the system responds accordingly. CMU researchers are working on a system that replaces the receiver with a camera and adds a programmable mirror to the emitter. These two components are programmed to sweep across an area, covering a plane instead of a single line. “In our test cell, we can read the joint position of the robot and then move the envelope around the robot as the joints move,” Aggarwal explains. Light curtains can also be used to track humans in space. CMU researchers are testing out the use of wireless communication in these systems, instead of ethernet cables, to make the system more scalable for large manufacturing plants.

From the Test Bed to the Shop Floor

Research from the test bed can make its way into the manufacturing industry in a variety of ways. MFI is co-located with the Advanced Robotics for Manufacturing (ARM) Institute, a Manufacturing USA manufacturing innovation institute that focuses on advancing American manufacturing through robotics and AI. ARM is a consortium with members in government, academia and industry. “Whatever we develop here, ARM is going make that available for all their members to test out their use cases,” Aggarwal explains. The idea is to produce pre-trained models, which manufacturers can customize for their own specific end uses. This is faster and easier than automating a task from scratch.

Companies have also spun out of MFI. Phlux Technology is working on commercializing the programmable light curtain technology. “They're working on getting it out to the market and getting certifications,” Aggarwal explains. “That is a tangible example of how our technologies get out into the world.”