Learning for Dynamic Subteaming and Voluntary Waiting in Heterogeneous Multi-Robot Collaborative Scheduling

Williard Joshua Jose and Hao Zhang
University of Massachusetts Amherst
IEEE International Conference on Robotics and Automation (ICRA) 2024
Best Paper Award Finalist on Multi-Robot Systems


Coordinating heterogeneous robots is essential for autonomous multi-robot teaming. To execute a set of dependent tasks as quickly as possible, and to complete tasks that cannot be addressed by individual robots, it is necessary to form subteams that can collaboratively finish the tasks. It is also advantageous for robots to wait for teammates and tasks to become available in order to form better subteams or reduce the overall completion time.

To enable both abilities, we introduce a new graph learning approach that formulates heterogeneous collaborative scheduling as a bipartite matching problem that maximizes a reward matrix learned via imitation learning. We design a novel graph attention transformer network (GATN) that represents the problem of collaborative scheduling as a bipartite graph, and integrates both local and global graph information to estimate the reward matrix using graph attention networks and transformers.

By relaxing the constraint of one-to-one correspondence in bipartite matching, our approach allows multiple robots to address the same task as a subteam. Our approach also enables voluntary waiting by introducing an idle task that the robots can select to wait. Experimental results have shown that our approach well addresses heterogeneous collaborative scheduling with dynamic subteam formation and voluntary waiting, and outperforms the previous and baseline methods.


Interpolate start reference image.

A motivating scenario for heterogeneous multi-robot collaborative scheduling with dynamic subteaming and voluntary waiting in an assembly application: Subteaming allows robots to dynamically build subteams to address tasks that cannot be completed by individual robots, while voluntary waiting enables robots to wait for additional robots and tasks to be available in order to form better subteams or reduce the overall task completion time.

Learning for Voluntary Waiting and Subteaming (LVWS)

Interpolate start reference image.

Overview of our LVWS approach: By maximizing a scheduling reward matrix R that is estimated using imitation learning, LVWS integrates GATs and transformers to learn a policy to determine robot-task assignments. LVWS enables subteaming by grouping multiple robots that collectively meet a task's requirements on capabilities and payload capacities. LVWS enables voluntary waiting by allowing robots to select the idle task tM+1.

Demo Videos

Simulation (5x speed)

We demonstrate our method in a manufacturing assembly case study run in a Gazebo simulation. There are 3 robot manipulators in 3 assembly cells, and there are also 3 ground vehicles with different payload capacities. The manufacturing process is decomposed into 14 tasks with varying capability and payload requirements.

We can see one of the jackal robots demonstrating voluntary waiting. We can also see the husky and jackal robot has formed a subteam to transport an oversized payload. These robots are autonomously operating according to the multi-robot task scheduling solution that was computed by LVWS. For this specific case study, the solution by LVWS is also the exact solution.

Real World (5x speed)

We can also demonstrate our method running on real-world robots. We have 2 human workers in assembly cells replacing the robot manipulators, and we have the same 3 ground vehicles. The simplified manufacturing process is decomposed into 7 tasks.

We also observed one of the jackals demonstrating voluntary waiting at the start. The two jackals also successfully transported an oversized payload by forming a subteam. The robots here are being teleoperated, but still according to the scheduling solution by LVWS, which is also the optimal solution for this case.


  author    = {Jose, Williard Joshua and Zhang, Hao},
  title     = {Learning for Dynamic Subteaming and Voluntary Waiting in Heterogeneous Multi-Robot Collaborative Scheduling},
  booktitle = {IEEE International Conference on Robotics and Automation},
  year      = {2024}