# Semantic Interaction in Augmented Reality Environments for Microsoft HoloLens

**Authors:** Peer Schüett, Max Schwarz, Sven Behnke  
**arXiv:** [2112.05846](https://arxiv.org/abs/2112.05846)  
**Date:** 2021-11-18  
**Summary page:** https://www.summarizepaper.com/en/paper/2112.05846v1/

## Key points

- Augmented Reality (AR) has potential for enhancing human-machine interaction in robotics
- AR provides a more intuitive and efficient way to display visualizations and receive user input in the environment where the system operates
- Microsoft HoloLens was used to capture indoor environments and display interaction cues with known object classes
- 3D mesh recorded by the HoloLens was annotated online with semantic classes using a projective approach, allowing for state-of-the-art 2D semantic segmentation methods to be used
- Results were fused onto the mesh, identifying prominent object segments that were displayed in 3D to the user, actions could be triggered by gesturing at objects
- Authors presented qualitative results and analyzed accuracy and performance of their method on an indoor dataset, provided examples of mapping runs showing how objects were detected and segmented in real-time using their approach
- To address potential issues with low resolution meshes during visualization of results, recommended showing only abstract icons as interaction cues instead of full meshes to visualize semantics
- Future work should generalize assumptions made to evaluate their impact on segmentation results; for example, assumptions regarding a static world and pre-processed mesh data could be re-evaluated by creating detailed server-sided meshes from raw data or improving semantic segmentation methods through synthetic training data that better fit real-world situations.
- Study highlights how AR can improve human-machine interaction in robotics by providing more intuitive ways to visualize and interact with systems in their environment.

## Summary

The use of Augmented Reality (AR) in robotics has shown great promise for enhancing human-machine interaction. In particular, AR can provide a more intuitive and efficient way to display visualizations and receive user input in the environment where the system operates. In this study, the Microsoft HoloLens was used to capture indoor environments and display interaction cues with known object classes. The 3D mesh recorded by the HoloLens was annotated online with semantic classes using a projective approach, allowing for state-of-the-art 2D semantic segmentation methods to be used. The results were then fused onto the mesh, identifying prominent object segments that were displayed in 3D to the user. Actions could be triggered by gesturing at the objects. The authors presented qualitative results and analyzed the accuracy and performance of their method on an indoor dataset. They also provided examples of mapping runs, showing how objects were detected and segmented in real-time using their approach. To address any potential issues with low resolution meshes during visualization of results, they recommended showing only abstract icons as interaction cues instead of full meshes to visualize semantics. In future work, it is suggested that some assumptions made should be generalized to evaluate their impact on segmentation results; for example, assumptions regarding a static world and pre-processed mesh data could be re-evaluated by creating detailed server-sided meshes from raw data or improving semantic segmentation methods through synthetic training data that better fit real-world situations. Overall, this study highlights how AR can improve human-machine interaction in robotics by providing more intuitive ways to visualize and interact with systems in their environment. It builds upon previous work such as Chen et al. 's framework for assigning semantic properties to real environments and demonstrates how mixed reality can enhance these properties in practical applications like first-person shooter apps. As AR technology continues to advance, it is likely that we will see even more innovative and effective ways to use it in robotics and beyond.