[Paper Review] An integrated dynamic method for allocating roles and planning tasks for mixed human-robot teams

Fabio Fusaro, Edoardo Lamon+2 more|arXiv (Cornell University)|Jan 1, 2021

Scheduling and Optimization Algorithms18 references25 citations

TL;DR

This paper proposes an integrated dynamic method using Behavior Trees (BTs) with online Mixed-Integer Linear Programming (MILP) sub-problems to allocate tasks and roles in real time within mixed human-robot teams. By modeling jobs as BTs with role allocation nodes that solve simplified MILP sub-problems based on action-worker costs—including task duration, availability, and ergonomics—the method achieves efficient, adaptive task planning with linear computational scaling and optimal agent assignment, validated through simulations showing reduced task completion times and improved workload distribution.

ABSTRACT

This paper proposes a novel dynamic method based on Behavior Trees (BTs) that integrates planning and allocation of tasks in mixed human robot teams, suitable for manufacturing environments. The Behavior Tree formulation allows encoding a single job as a compound of different tasks with temporal and logic constraints. In this way, instead of formulating an offline centralized optimization problem, the role allocation problem is solved with multiple simplified online optimization sub-problems, without complex and cross-schedule task dependencies. These sub-problems are defined as Mixed-Integer Linear Programs (MILPs), that, according to the worker-actions related costs and the workers' availability, allocate the yet-to-execute tasks among the available workers. To characterize the behavior of the developed method, we opted to perform different simulation experiments, in which the results of the action-worker allocation and the computational complexity are evaluated. The obtained results, due to the nature of the algorithm and to the possibility of simulating the agents' behavior, illustrate adequately also how the algorithm performs in real experiments.

Motivation & Objective

• . The paper addresses the challenge of dynamically allocating roles and planning tasks in mixed human-robot teams for flexible manufacturing.
• It aims to overcome the limitations of traditional two-phase approaches (offline planning and centralized optimization) by integrating role allocation and task planning into a single, reactive framework.
• The objective is to reduce computational complexity while maintaining optimal task assignment by solving only subsets of tasks through online MILP sub-problems.
• The method aims to improve team efficiency by considering agent-specific factors such as availability, task duration, expertise, and ergonomics in cost functions.
• The goal is to enable real-time adaptation to dynamic changes in team composition and task constraints, ensuring temporal and logical dependencies are preserved.

Proposed method

• . The method uses Behavior Trees (BTs) to model job-level behavior, where the root node controls the execution of all tasks, and each task is managed via a Role Allocator node.
• Each Role Allocator node dynamically solves a Mixed-Integer Linear Program (MILP) to assign the next set of actions to the most suitable agent based on action-worker costs.
• The action-worker cost function incorporates task duration, agent availability, and ergonomic factors, with three variants: binary availability, continuous availability cost (Equation 4), and zero availability cost.
• The BT architecture includes four custom nodes: Role Allocator (solves MILP), Agent Handler (routes actions to human or robot), Human Communication (notifies human workers), and Action (executes robot tasks).
• The system is implemented in C++ using ROS Melodic and the BehaviorTree.CPP library, with Osi/GLPK for MILP solving.
• The method decouples the centralized planning problem into multiple lightweight, online sub-problems, reducing computational burden while preserving task constraints.

Experimental results

Research questions

• RQ1 . How can task allocation and planning be integrated in a single dynamic framework for mixed human-robot teams to reduce computational complexity?
• RQ2 What impact does the definition of agent-related costs—particularly availability—have on the quality of task and role allocation?
• RQ3 How does the computational time scale with increasing numbers of tasks and agents in the system?
• RQ4 Can the method achieve optimal task assignment while minimizing overall job completion time and agent waiting times?
• RQ5 How does the method perform under varying levels of task parallelism and agent availability?

Key findings

• . The computational time for the entire framework scales approximately linearly with the number of sequential and parallel tasks, indicating good scalability.
• Solving a single large MILP of size N is faster than solving N smaller MILPs, even though the latter involves more sub-problems, due to reduced overhead and better solver utilization.
• As the number of workers increases, the computation time decreases until the maximum degree of parallelism is reached (at 15 workers), after which it remains approximately constant.
• The use of a continuous availability cost (Equation 4) leads to better task allocation than binary or zero availability costs, as it minimizes both task execution time and waiting time for subsequent tasks.
• In the simulation with 14 actions, the method assigned action a12 to worker w1 (minimizing total time) when using the continuous availability cost, outperforming both binary and zero-cost variants.
• The Gantt chart analysis confirmed that the continuous availability cost model reduced idle time and improved coordination, especially for actions requiring sequential execution after parallel task sets.