Photo credit: www.therobotreport.com
Recent research from the University of Massachusetts Amherst suggests that utilizing teams of self-organizing robots may be more advantageous in industrial settings than traditional multipurpose robotic systems. These collaborative teams are capable of completing tasks that could be monotonous or risky for limited human workers, although they typically require pre-programming or central control to function effectively. The UMass Amherst team discovered that when robots are programmed to form their own teams and wait for one another voluntarily, they can achieve faster task execution.
“The ongoing debate centers around whether to develop a single, versatile humanoid robot that can handle various tasks or to rely on a coordinated group of robots,” explained Hao Zhang, a co-author of the study, who is an associate professor with the Manning College of Information and Computer Sciences at UMass Amherst. He also leads the Human-Centered Robotics Lab.
According to Zhang, deploying a team of robots in manufacturing scenarios can be more cost-effective as it allows for enhanced utilization of each robot’s capabilities. The primary challenge lies in the coordination of these diverse robotic entities, which may vary in mobility, functionality, and task capabilities.
The research team proposed a unique scheduling method termed “learning for voluntary waiting and sub-teaming” (LVWS), which they believe could significantly enhance automation processes in sectors such as manufacturing, warehousing, and agriculture.
The findings from this research earned recognition as finalists for the Best Paper Award on Multi-Robot Systems at the upcoming IEEE International Conference on Robotics and Automation 2024.
Evaluating the LVWS Method
To assess their robot orchestration strategy, the researchers implemented a computer simulation involving six robots completing 18 distinct tasks. The performance of the LVWS method was compared to four other robotic orchestration approaches. The team’s computer model was designed with a perfect solution known for accomplishing the tasks in the shortest time possible.
During the simulation, researchers measured the efficiency of each model against this perfect solution to determine their suboptimality. The alternative approaches displayed suboptimality ranging from 11.8% to 23%, while the LVWS method registered only 0.8% suboptimality.
The researchers clarified how allowing a robot to wait might enhance overall team performance by presenting a scenario involving three robots: two capable of lifting 4 lb. each, and a third designed to lift 10 lb. If one of the smaller robots is occupied, it could be more strategic for the larger robot to wait for the second smaller robot to assist with moving a 7-lb. box. In this case, the larger robot could then focus on a more suitable task, optimizing the use of their combined resources.
Rationale Behind LVWS Versus Exact Solutions
While the researchers used a perfect solution as a benchmark, they acknowledged that this level of optimization is impractical for real-world scenarios involving robotic lifting tasks.
“The challenge of relying on an exact solution is that computing it can be extremely time-consuming,” noted Jose. “As the number of robots increases, the complexity grows exponentially, making it impossible to retrieve an optimal solution in a timely manner.”
In simulations with 100 tasks, where deriving a precise solution becomes unmanageable, the team’s LVWS approach completed the tasks in 22 time steps, compared to 23.05 to 25.85 time steps taken by the alternative methods. Such efficiency gains can have substantial implications in production environments.
Zhang expressed aspirations for this research to propel advancements in robot teaming, particularly in relation to scalability. He pointed out that while a single humanoid robot might suit the confines of a home, multi-robot systems stand out as more effective solutions in expansive industrial settings that necessitate specialized operations.
Source
www.therobotreport.com