New research from the University of Massachusetts Amherst has demonstrated that programming robots to create their own teams and voluntarily wait for their teammates can lead to faster completion of tasks, offering potential benefits for industries such as manufacturing, agriculture, and warehouse automation. The study, which was recognized as a finalist for the Best Paper Award on Multi-Robot Systems at the IEEE International Conference on Robotics and Automation 2024, was led by Hao Zhang, an associate professor at UMass Amherst’s Manning College of Information and Computer Sciences. Zhang and his team developed a learning-based approach for scheduling robots, known as learning for voluntary waiting and subteaming (LVWS).
In a manufacturing setting, using a team of robots can be more cost-effective as it allows for the maximization of each robot’s capabilities. However, coordinating a diverse set of robots with varying strengths and limitations can be a challenge. The LVWS approach created by Zhang and his team aims to address this issue by enabling robots to work collaboratively on tasks that require multiple robots, such as moving a large box that cannot be carried by a single robot. In addition, the approach incorporates the concept of voluntary waiting, allowing robots to wait for optimal conditions to perform tasks, rather than always carrying out smaller tasks immediately.
To test the effectiveness of their LVWS approach, the researchers conducted simulations with six robots and 18 tasks, comparing the results with four other methods. The LVWS method was found to be 0.8% suboptimal, significantly outperforming the other comparison methods, which ranged from 11.8% to 23% suboptimal. This indicates that the LVWS approach can lead to task completion closer to the best possible or theoretical solution, making the team of robots more efficient in completing tasks.
The researchers also investigated the impact of making robots wait on the overall speed of task completion. By considering scenarios where robots have different capabilities, such as lifting capacities, the study demonstrated that having robots wait for optimal team configurations can lead to faster completion of tasks. Despite the possibility of calculating an optimal solution, the researchers highlighted that the computational time required to do so would be impractical, particularly for larger numbers of robots and tasks. Therefore, the use of a scheduler like LVWS becomes essential in optimizing task allocation and coordination in multi-robot systems.
Looking ahead, Zhang aims for this research to contribute to the advancement of automated robot teams, especially as scalability becomes a key factor. While a single, humanoid robot may be suitable for small environments like single-family homes, multi-robot systems offer more advantages in large industrial settings that require specialized tasks. The research was funded by the DARPA Director’s Fellowship and a U.S. National Science Foundation CAREER Award, highlighting the significance of the findings in advancing the field of robotics and automation.