# Incremental 3D Scene Completion for Safe and Efficient Exploration Mapping and Planning

**Authors:** Lukas Schmid, Mansoor Nasir Cheema, Victor Reijgwart, Roland Siegwart, Federico Tombari, Cesar Cadena  
**arXiv:** [2208.08307](https://arxiv.org/abs/2208.08307)  
**Date:** 2022-08-17  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2208.08307v1/

## Key points

- Paper title: "Incremental 3D Scene Completion for Safe and Efficient Exploration Mapping and Planning"
- Authors: Lukas Schmid, Mansoor Nasir Cheema, Victor Reijgwart, Roland Siegwart, Federico Tombari, Cesar Cadena
- Challenge addressed: Exploring unknown environments in robotics
- Proposed approach: SC-Explorer integrates deep learning into exploration by leveraging 3D scene completion for informed, safe, and interpretable mapping and planning
- SC-Explorer combines scene completion using an incremental fusion mechanism with a hierarchical multi-layer mapping approach
- Introduces an informative path planning method that takes advantage of their mapping approach and a scene-completion-aware information gain strategy
- Tested in the context of a Micro Aerial Vehicle (MAV) through high-fidelity simulation experiments using only mobile hardware
- Results show SC-Explorer can accelerate coverage of an environment by 73% compared to baseline methods with minimal impact on map accuracy
- Scene completions guide the robot to choose more informative paths, speeding up the measurement of the scene with the robot's sensors by 35%
- Availability: Authors make their methods available as open-source code on GitHub

## Summary

The paper "Incremental 3D Scene Completion for Safe and Efficient Exploration Mapping and Planning" by Lukas Schmid, Mansoor Nasir Cheema, Victor Reijgwart, Roland Siegwart, Federico Tombari, and Cesar Cadena addresses the challenge of exploring unknown environments in robotics. The authors propose a novel approach called SC-Explorer that integrates deep learning into exploration by leveraging 3D scene completion for informed, safe, and interpretable mapping and planning. <br/>
SC-Explorer combines scene completion using an incremental fusion mechanism with a hierarchical multi-layer mapping approach to ensure the safety and efficiency of the robot during exploration. The authors also introduce an informative path planning method that takes advantage of their mapping approach and a scene-completion-aware information gain strategy. <br/>
While the method is generally applicable, it is evaluated in the context of a Micro Aerial Vehicle (MAV) through high-fidelity simulation experiments using only mobile hardware. The results show that SC-Explorer can accelerate coverage of an environment by 73% compared to baseline methods with minimal impact on map accuracy. Additionally, even if scene completions are not included in the final map, they can guide the robot to choose more informative paths, speeding up the measurement of the scene with the robot's sensors by 35%. <br/>
The authors make their methods available as open-source code on GitHub. Overall, this paper presents a comprehensive framework for safe and efficient exploration mapping and planning in unknown environments using deep learning-based scene completion. The proposed approach shows promising results in improving exploration efficiency while maintaining safety considerations for autonomous systems.