ARENA: Adaptive Risk-aware and Energy-efficient NAvigation for Multi-Objective 3D Infrastructure Inspection with a UAV

TL;DR

ARENA addresses multi-objective UAV navigation in complex 3D environments by introducing an online, risk-aware framework that jointly optimizes safety, time, and energy. It uses a 4D NURBS trajectory representation to smoothly encode position and velocity, solves for a Pareto front with NSGA-II, and selects trajectories via a real-time voting scheme driven by mission risks such as wind, localization accuracy, communications, and battery state. Key contributions include a novel risk-based voting mechanism, a 4th velocity dimension in the trajectory representation, a semi-empirical energy model, and validation through simulations and real-world power-line inspections showing robust adaptation and 14% worst-case power-model error. The approach enhances UAV autonomy and reliability in evolving 3D mission contexts, with potential applicability to diverse infrastructure inspection tasks beyond power lines, and lays groundwork for real-time, energy-aware, risk-sensitive autonomous navigation.

Abstract

Autonomous robotic inspection missions require balancing multiple conflicting objectives while navigating near costly obstacles. Current multi-objective path planning (MOPP) methods struggle to adapt to evolving risks like localization errors, weather, battery state, and communication issues. This letter presents an Adaptive Risk-aware and Energy-efficient NAvigation (ARENA) MOPP approach for UAVs in complex 3D environments. Our method enables online trajectory adaptation by optimizing safety, time, and energy using 4D NURBS representation and a genetic-based algorithm to generate the Pareto front. A novel risk-aware voting algorithm ensures adaptivity. Simulations and real-world tests demonstrate the planner's ability to produce diverse, optimized trajectories covering 95% or more of the range defined by single-objective benchmarks and its ability to estimate power consumption with a mean error representing 14% of the full power range. The ARENA framework enhances UAV autonomy and reliability in critical, evolving 3D missions.

Figures (7)

• Figure 1: Representation of every non-dominated trajectory at the end of ARENA's process. An RGB model depicts the various objectives' influence on a trajectory.
• Figure 2: Voting algorithm diagram
• Figure 3: Path comparison and Gazebo visualization: (a)-(c) Path comparison for different risk sets, with an RGB model indicating the risk influencing trajectory selection: red for low battery, green for localization, and blue for high wind risk. Obstacles are shown using an Octomap. (b)-(d) Gazebo simulations of a construction site and power lines model.
• Figure 4: Sensitivity study of the voting algorithm. The hyperparameters are set as follows: $v_{max}=2.0$, $a_{max}=2.2$, $p=3$, $r_{sdf_{min}}=1$, $r_{sdf{max}}=5$, $r_{ch_{max}}=2$, $\delta_{rope}=5.0$, $N_{gen}=1000$, $N_{pop}=40$, $N_{nurbs}=50$. Every coefficients set is evaluated over 3 iterations.
• Figure 5: Metric sensitivity to risk factors: average distance to obstacles, path duration, and energy consumption.