After much trial and error, Jacobi’s founders, including roboticist Ken Goldberg, say they’ve cracked it. Their software, based on research by a paper they published in Robotics Science in 2020, it is designed to work with the top four robotic palletizing arm manufacturers. It uses deep learning to create a “first plan” of how an arm might move an object on the pallet. It then uses more traditional robotics methods, such as optimization, to check whether the movement can be made safely and without malfunctions.
Jacobi aims to replace the legacy methods that customers currently use to train their bots. In the conventional approach, robots are programmed using tools called “teaching pendants” and customers usually have to manually guide the robot to show it how to pick up each individual box and place it on the pallet. The entire coding process can take months. Jacobi says the AI-based solution promises to reduce that time to a day and can calculate movements in less than a millisecond. The company says it plans to launch its product later this month.
Billions of dollars are being poured into AI robotics, but most of the excitement is geared toward next-generation robots that promise to be capable of many different tasks—like the humanoid robot which helped the Figure raise $675 million from investors including Microsoft and OpenAI, reaching a $2.6 billion valuation in February. Against this backdrop, using artificial intelligence to train a better box-stacking robot might seem pretty basic.
Indeed, Jacobi’s seed funding round pales in comparison: $5 million led by Moxxie Ventures. But amid hype surrounding promising innovations in robotics that could take years to materialize, palletizing may be the warehouse problem that AI is best poised to solve in the near term.
“We have a very pragmatic approach,” says Max Cao, co-founder and CEO of Jacobi. “These jobs are doable and we can get a lot of adoption in a short amount of time, compared to some of the moon shots out there.”
Jacobi’s software product includes a virtual studio where customers can create copies of their setups, noting factors such as which robot models they have, what types of boxes will come off the conveyor belt and which direction the labels should face. A warehouse that carries sporting goods, say, can use the program to find the best way to stack a mixed pallet of tennis balls, rackets and clothing. Jacobi’s algorithms will then automatically design the many movements the robotic arm must make to stack the pallet, and the instructions will be transmitted to the robot.
JACOBI ROBOTICS
The approach combines the benefits of fast computation provided by artificial intelligence with the precision of more traditional robotics techniques, says Dmitri Berenson, a professor of robotics at the University of Michigan who is not involved with the company.
“They’re doing something very sensible here,” he says. Much modern robotics research is betting big on artificial intelligence, hoping that deep learning can augment or replace more manual training by having the robot learn from previous examples of a given movement or task. But by ensuring that the predictions produced by deep learning are checked against the results of more traditional methods, Jacobi is developing design algorithms that will likely be less prone to error, Berenson says.
The design speed that could result “pushes it into a new category,” he adds. “You won’t even notice the time it takes to calculate a move. This is very important in the industrial environment, where every pause means delays.”