Multi-Objective Planning for Autonomous Robots

Autonomous robots often operate in environments where navigation involves multiple, potentially conflicting objectives. Traditional path planning methods, such as those used in the ROS Navigation2 (Nav2) stack, optimize a single additive cost function. However, many real-world problems require reasoning over hierarchically structured and non-comparable objectives—for example, prioritizing safety above efficiency, or law compliance above travel time. To address these limitations, this project implements a custom multi-objective path planning plugin for the Navigation2 framework, based on the theoretical framework presented in “Hierarchical Multiobjective Shortest Path Problems” by Slutsky et al. (2020). 

The referenced paper introduces a graph-based algorithm that extends the classical shortest path problem to handle multiobjective cost structures represented as ordered combinations of algebraic cost monoids. This formulation allows for objectives that are not easily merged into a single additive cost, while maintaining computational efficiency through an iterative graph search procedure. Building upon this foundation, the project develops a custom Nav2 planner plugin that integrates this hierarchical multiobjective planning algorithm into both simulated and physical ROSBot platforms. The system enables autonomous robots to compute paths that respect objective hierarchies, adapt dynamically to environmental changes, and maintain optimal performance across competing goals. 

By bridging the theoretical base established from Slutsky et al. with the modularized robotics infrastructure of Nav2, this project demonstrates a framework for multi-objective navigation. The resulting planner contributes to advancing research in multi-objective decision making autonomous systems, and further applications in different scenarios such as drones.

Committee: Simanta Mitra (major professor)