Mujin Inc., a Japanese company that has brought “machine intelligence” to robots across Asia since 2011, today opened its first office in North America. Mujin Corp.'s facility in Sandy Springs, Ga., north of Atlanta, will house its expanding engineering, sales, and support staff, including Ross Diankov, co-founder of Mujin and CEO of Mujin Corp.
“Companies that want to automate mundane and repetitive material-handling tasks face a myriad of challenges, from the high costs of developing solutions for their difficult applications to unscheduled downtime and reprogramming costs when things don’t go as planned, or when robots must be reprogrammed due to a change in product or workflow,” stated Diankov.
“As some of the largest companies in Asia have experienced, Mujin will bring a new wave of robotics technology to the U.S. market, with robots no longer needing to be taught how to move explicitly,” he claimed. “Instead, the robots will already ‘know’ what they need through what we call ‘machine intelligence,’ which enables more capability and efficiency for robot picking applications that were previously impractical or difficult to deploy.”
Tokyo-based Mujin said it has designed a common platform for industrial and collaborative robot arms specializing in logistics and other pick-and-place applications.
The Mujin Controller uses real-time perception, motion planning, simulation, and universal control to enable robots to handle complex tasks without the need for coding, the company said. It added that it has worked to make automation easier to deploy, more accurate and reliable, and less costly.
Mujin minimizes human intervention
“As we come into the Americas, we've been talking to end users and integrators,” Diankov told Robotics 24/7. “When we ask them how many systems are in production, they're still in the single digits. Amazon can hire tens of thousands of people and use thousands of robots, but very few companies can do that.”
“We're seeing human beings picking everywhere,” he said. “Most facilities start with pilot projects, but a lot of automation deployments end there because the solution provider wasn't at the technology level to be stable in production.”
“Having systems stop all the time kills productivity gains. You shouldn't need a human being to watch a robot all the time,” said Diankov. “Mujin has been doing this for the past five years.”
“One of our recent deployments included 20 robots at one warehouse. Each of them collectively does 100,000 picks, 23 hours per day and stops once or twice per day,” he said. “At that level, you're getting real value out of automation.”
Machine intelligence vs. machine learning
Conventional robots require experts to program each movement or use machine learning to learn how to pick items over time. The Mujin Controller takes a different approach, allowing for more autonomous and production-capable applications, claimed the company.
“Artificial intelligence companies gather data, close their eyes, and cross their fingers, hoping a model is created to get to 80% to 90% success in a production scenario,” said Diankov. “One customer in Japan had a mixed-SKU depalletizing system. Whenever the robot couldn't pick, the company would have to send the data to the U.S. and then wait two to four days to retrain the model. It stopped production.”
“With AI, you could tweak a random parameter and think a problem is fixed, but its root cause could be hidden,” he explained. “Robots should be deterministic and reliable. You take a train from Point A to Point B, and if the train breaks, you can identify the cause to fix it. We call the base concept of our deterministic way of picking machine intelligence.”
With Mujin Controller, users begin by modeling the environment and setting relationships between the robots and target objects at a high level. The system then allows the robot to safely perform tasks by offering high-level goals without explicitly telling the robot where to go or how to move.
“We're measuring the properties of the physical world for a better model,” Diankov said. “Then we can go pick an item in the best way possible. It's not just CAD or a point cloud, but also weight, material, and force. We've built these things, which are quality-assured and verified. Machine intelligence comes from 'measurement intelligence.'”
“With the robot motion now computed in real time without human intervention, the system must understand the intuitions behind completing tasks and then position them in a way that enables the robot to dynamically adapt to changing circumstances,” he added. “Mujin makes every robot more capable, efficient, and reliable.”
Mujin makes both hardware and software
“One important distinction between Mujin and other companies in this space is that we make hardware such as grippers,” said Diankov. “Most of the time, the systems design comes from us, and an integrator implements it.”
“In Japan and China, companies are more willing to start from a blank slate for entire system throughput—they need in and out lines, rejects, AGVs [automated guided vehicles], etc.,” he said. “We can design the whole warehouse with integrators and operators. No other country goes to that level.”
“A robot by itself is dumb and is useful only when you connect it upstream and downstream,” Diankov asserted. “Customers have to spend more money localizing projects, connecting sensors, making sure of assumptions for picking, or managing SKU size. This is not just for pick and drop, but also palletizing, materials handling, and industrial automation.”
Mujin to show robots at Pack Expo
Mujin plans to showcase palletizing with autonomous mobile robots (AMRs) and its award-winning technology at Booth SU-7233 at Pack Expo in Las Vegas from Sept. 27 to 29. The demonstration will include a Mujin gripper capable of handling cases, slip sheets, and pallets to build multiple pallets into a single stack.
Once the palletizing operation is complete, the robots will deliver the full pallet to the pallet breakdown area. The AMRs will also bring empty pallets and slip sheets to the workcell. A robot arm equipped with 3D vision on the gripper will find the position of the pallets and slip sheets to pick.
If a depalletizing system doesn't need localization but can still pick everything without singulation on the back end, it could eliminate the need for singulation, allowing for smaller conveyors and robot cells, said Mujin.
“We have the world's most reliable mixed depalletizer,” Diankov said. “When we see something unkown, we have a way to measure and verify every single box we pick, so the robot never gets stuck.”
“There's a huge gap between a demonstration and production,” he acknowledged. “We've started projects and going after U.S. customers. They can come in today, and have have a lot of resources to service any customer, no matter the size.”