New research from the University of Massachusetts Amherst shows that programming robots to form their own teams and volunteer to wait on their teammates leads to faster task completion with the potential to improve manufacturing, agriculture and warehouse automation. It happens. This research was recognized as a finalist for the Best Paper Award on Multi-Robot Systems at the IEEE International Conference on Robotics and Automation 2024.
Associate Professor Hao Zhang, one of the study's authors, says, “There is a long history of debate about whether we want to build a single, powerful humanoid robot that can do all the work, or whether we want to have a collection of robots. There is a team that can cooperate.” at the UMass Amherst Manning College of Information and Computer Sciences and director of the Human-Centered Robotics Lab.
In a manufacturing setting, a robot team can be less expensive because it maximizes the potential of each robot. The challenge then becomes: How do you coordinate a diverse set of robots? Some may be fixed in place, some mobile; Some can lift heavy materials, while others are suitable for smaller tasks.
As a solution, Zhang and his team created a learning-based approach to scheduling robots called Learning for Voluntary Waiting and Subteaming (LVWS).
“Robots have big jobs, just like humans,” says Zhang. “For example, they have a big box that a robot can't carry.
The second behavior is voluntary waiting. “We want the robot to be able to proactively wait because, if it always chooses greedy solutions to perform smaller tasks that are immediately available, sometimes the larger task is never executed. will go,” Zhang explains.
To test their LVWS approach, they assigned 18 tasks to six robots in a computer simulation and compared their LVWS approach to four other approaches. In this computer model, there is a known, optimal solution to complete the scenario in the fastest time. The researchers ran different models through simulations and calculated how poorly each method performed compared to the perfect solution, a measure known as suboptimal capacity.
The compared methods ranged from 11.8% to 23% suboptimal. The new LVWS method was the best at 0.8%. “So this solution is close to the best possible or theoretical solution,” says Willard Jose, an author on the paper and a doctoral student in computer science in the Human-Centered Robotics Lab.
How does making a robot wait make the whole team faster? Consider this scenario: You have three robots — two that can lift four pounds and one that can lift 10 pounds. One of the smaller robots is busy with a different task and has a seven-pound box that needs to be moved.
“Instead of having the big robot perform the task, it would be more beneficial for the small robot to wait for the other small robot to perform the large task together because the large robot's resource is being used by another large robot. Better suited to do the job.” says Jose.
If it is possible to determine an optimal response first, why do robots even need a scheduler? “The problem with using this exact solution is that it takes a lot of time,” Joss explains. “With a large number of robots and tasks, it's worth it. You can't find the best solution in a reasonable amount of time.”
Looking at models using 100 tasks, where it is difficult to calculate an exact solution, they found that their method completed the task in 22 time steps, compared to 23.05 to 25.85 time steps for comparable models. .
Zhang hopes the work will further help these teams develop automated robots, especially when the question of scale comes up. For example, he says, a single, humanoid robot might fit better into the small footprint of a single-family home, while multi-robot systems are better options for larger industrial environments that require specialized tasks.
This research was funded by a DARPA Director's Fellowship and a US National Science Foundation CAREER award.