Researchers at the University of Massachusetts Amherst have conducted a new study revealing that programming robots to autonomously form teams and patiently wait for their colleagues leads to faster task completion times, with far-reaching implications for industries such as manufacturing, agriculture, and warehouse automation. Our analysis received recognition as a finalist for the prestigious Greatest Paper Award in the field of Multi-Robotic Systems at the 2024 IEEE International Conference on Robotics and Automation.
According to Hao Zhang, an associate professor in the University of Massachusetts Amherst’s Manning School of Information and Computer Sciences and director of the Human-Centered Robotics Lab, there is a long-standing debate about whether to create a single, highly capable humanoid robot that can perform all tasks or instead develop a team of robots that can collaborate effectively.
In a production setting, a robotic crew will prove cost-effective due to its ability to maximize the potential of each robot. How do you effectively orchestrate a diverse array of robots to work in tandem? Vessels of varying types and capabilities exist, with some secured firmly in position while others are designed with internal compartments; some are capable of hauling substantial cargo, whereas others are better suited for lighter tasks.
Zhang and his team developed a learning-based approach to scheduling robots called Studying For Voluntary Ready And Sub-Teaming (LVWS), which streamlined the process.
According to Zhang, robots’ responsibilities are comparable to those of humans. For example, they have a large field that is too vast to be covered by a single robot. The scenario requires a certain quantity of robots to collaborate in accomplishing the task.
Inconsistent behaviour is intentionally prepared. “We want robots to possess the capacity to pause proactively when they consistently choose a grasping solution to accomplish smaller tasks that are readily available, thereby preventing larger projects from being initiated.”
To evaluate the efficacy of their LVWS approach, researchers conducted a PC-based simulation, assigning 18 tasks to six robots and benchmarking it against four alternative strategies. On this laptop model, there is a well-established solution for resolving the issue in the shortest possible timeframe. Researchers simulated various approaches, calculating the suboptimality, or degree of deviation from the optimal solution, for each method.
The comparability strategies exhibited a significant variability, ranging from 11.8% to 23% below optimal levels. While the LVWS technique showed promise, its initial iteration was approximately 0.8% short of optimal performance. Williard Jose, a doctoral student in computer science at the Human-Centered Robotics Lab, notes that the optimal solution lies close to the theoretical maximum.
The delay in making a robotic wait significantly reduces the overall processing time of the entire system by allowing for more efficient task management and streamlining of tasks. Consider the scenario where you possess three robots, comprising two with a load capacity of four kilograms each and one capable of handling ten kilograms. A lone small robot toils away, tasked with relocating a massive 7-pound field.
“As an alternative to relying on a single, large robot to accomplish the task, it would be more practical for multiple smaller robots to work together, with each one focusing on its own specific role. This collaborative approach would allow the larger robot to dedicate its resources to a different, equally important task.”
If it’s inherently possible to identify the most effective response from the start, then what’s driving robots to seek a scheduling mechanism in the first instance? According to Jose, computing the precise answer poses significant difficulties due to its extremely lengthy calculation time. As robot populations grow and responsibilities expand, the impact becomes exponentially more significant. It’s unlikely that you’ll achieve the best possible outcome within a limited timeframe.
Upon examining fashion’s utilization of 100 tasks, researchers found that when it was impractical to calculate a precise answer, their approach completed the tasks in 22 timesteps, outperforming the comparative models with estimated ranges of 23.05 to 25.85 timesteps.
Zhang aims to further advance the development of automated robotics groups, particularly in scenarios where scalability is a crucial factor. For instance, he posits that a solitary, humanoid robot could prove an even more suitable fit within the compact confines of a single-family residence, whereas multi-robot systems would be a superior option for a large-scale industrial setting demanding specialized tasks.
The analysis was funded through the auspices of the Defense Advanced Research Projects Agency’s (DARPA) esteemed Director’s Fellowship, as well as a grant from the United States government. Nationwide Science Basis CAREER Award.
