# Perception-and-Energy-aware Motion Planning for UAV using Learning-based Model under Heteroscedastic Uncertainty

**Authors:** Reiya Takemura, Genya Ishigami  
**arXiv:** [2309.14272](https://arxiv.org/abs/2309.14272)  
**Date:** 2023-09-25  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2309.14272v1/

## Key points

- Study focuses on perception-and-energy-aware motion planning for UAVs in GNSS-denied environments
- Proposed planner optimizes a cost function considering energy consumption and LiDAR perception quality
- High-fidelity simulator used to gather flight dataset for learning energy consumption patterns and uncertainty in LiDAR measurements
- Models designed to account for variations in horizontal velocity for accurate estimation of energy usage and perception quality
- Planner aims to reduce battery consumption while minimizing localization errors during navigation
- Simulation experiments validate efficacy of planner in balancing energy efficiency with perception quality under uncertainty
- Research submitted for presentation at 2024 IEEE ICRA, offering insights into improving motion planning strategies for UAVs in GNSS-denied conditions
- Authors have shared open-source code for further exploration and development at https://gitlab.com/ReI08/perception-energy-planner

## Summary

, , , , 

Authors Reiya Takemura and Genya Ishigami present a study on perception-and-energy-aware motion planning for unmanned aerial vehicles (UAVs) operating in GNSS-denied environments. In these challenging conditions, UAVs must navigate efficiently and reliably without the aid of GNSS signals. The proposed planner addresses this by optimizing a cost function that considers both the total energy consumption of the UAV and the quality of perception provided by a Light Detection and Ranging (LiDAR) sensor mounted on the UAV. To enhance its effectiveness, a high-fidelity simulator is used to gather a flight dataset, which is then utilized to learn about energy consumption patterns and heteroscedastic uncertainty associated with LiDAR measurements. These models are specifically designed to account for variations in horizontal velocity, enabling the online planner to accurately estimate energy usage and perception quality. By doing so, the planner aims to reduce battery consumption while minimizing localization errors during navigation. Simulation experiments conducted in a photorealistic environment validate the efficacy of the proposed planner in balancing energy efficiency with perception quality under heteroscedastic uncertainty. The results demonstrate that by incorporating learning-based models and effectively addressing uncertainties, UAVs can navigate GNSS-denied environments more efficiently and reliably. This research was submitted for presentation at the 2024 IEEE International Conference on Robotics and Automation (ICRA), providing valuable insights into improving motion planning strategies for UAVs operating in challenging GNSS-denied conditions. Additionally, the authors have made their open-source code available for further exploration and development at https://gitlab.com/ReI08/perception-energy-planner.