Researchers develop new method for successful collaboration between robots

New research from the University of Massachusetts Amherst shows that robots programmed to form their own teams and voluntarily wait for their teammates can complete tasks faster. This could improve automation in manufacturing, agriculture and warehousing. The study was published in 2024 IEEE International Conference on Robotics and Automation (ICRA).

This research was selected as a finalist for the Best Paper Award for Multi-Robot Systems at the IEEE International Conference on Robotics and Automation 2024.

“There has long been a debate about whether we want to build a single, powerful humanoid robot that can do all the tasks, or whether we want to have a team of robots that can work together,” says one of the study’s authors, Hao Zhang, associate professor at the UMass Amherst Manning College of Information and Computer Sciences and director of the Human-Centered Robotics Lab.

In a manufacturing environment, a robot team can be more cost-effective because it maximizes the capabilities of each individual robot. The challenge then becomes how to coordinate a multitude of robots. Some may be fixed, others mobile; some can lift heavy materials, while others are suited to smaller tasks.

As a solution, Zhang and his team developed a learning-based approach to robot planning called “Learning for Voluntary Waiting and Subteaming” (LVWS).

“Robots have big tasks, just like humans,” says Zhang. “For example, you have a big box that cannot be carried by a single robot. In this scenario, multiple robots are needed to work together.”

The other behavior is voluntary waiting. “We want the robot to be able to actively wait, because if it simply chooses a greedy solution to perform smaller and smaller tasks that are immediately available, sometimes the larger task will never be performed,” explains Zhang.

To test their LVWS approach, they gave six robots 18 tasks in a computer simulation and compared their LVWS approach with four other methods.

In this computer model, there is a known, perfect solution to complete the scenario in the shortest amount of time. The researchers ran the different models through the simulation and calculated how much worse each method performed compared to that perfect solution, a measure called suboptimality.

The comparison methods were between 11.8% and 23% suboptimal. The new LVWS method was 0.8% suboptimal. “So the solution is close to the best possible or theoretical solution,” says Williard Jose, author of the paper and a doctoral student in computer science at the Human-Centered Robotics Lab.

How can having one robot wait make the whole team faster? Consider this scenario: You have three robots – two that can lift 4 pounds each and one that can lift 10 pounds. One of the small robots is busy with another task and there is a seven-pound box that needs to be moved.

“Instead of the big robot performing this task, it would be more beneficial if the small robot waited for the other small robot and then they completed the big task together, because the larger robot’s resource is better suited to another big task,” says Jose.

If it’s even possible to find an optimal answer, why do robots need a planner at all? “The problem with using that exact solution is that it takes a very long time,” Jose explains. “With larger numbers of robots and tasks, it’s exponential. You can’t find the optimal solution in a reasonable amount of time.”

When looking at models with 100 tasks where it is impossible to compute an exact solution, they found that their method completed the tasks in 22 time steps, compared to 23.05 to 25.85 time steps for the comparison models.

Zhang hopes this work will help advance the progress of these teams of automated robots, especially when the question of scale comes into play.

For example, he says a single humanoid robot may fit better in the small space of a single-family home, while multi-robot systems may be a better choice for a large industrial environment that requires specialized tasks.