Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Bayesian Deep Learning for Segmentation for Autonomous Safe Planetary Landing

Created by
  • Haebom

Author

Kento Tomita, Katherine A. Skinner, Koki Ho

Outline

This paper presents a Bayesian deep learning-based segmentation method for risk detection for autonomous planetary landings. Conventional computer vision-based methods suffer from performance degradation as sensor noise increases. In this paper, we utilize Bayesian deep learning to simultaneously generate a safety prediction map and its uncertainty map. We then propose a reliable method for identifying safe landing sites by filtering out uncertain pixels using the uncertainty map. Using simulated data from the Mars HiRISE digital terrain model, we experimentally validate the performance of the proposed method under various uncertainty thresholds and noise levels.

Takeaways, Limitations

•
Takeaways:
◦
We present a safe landing point detection method that enhances the reliability of safety predictions by utilizing Bayesian deep learning.
◦
Improve safety by eliminating uncertain predictions through uncertainty maps.
◦
Demonstrates robust risk factor detection performance against sensor noise.
•
Limitations:
◦
Performance verification in actual mission environments is required through experiments using simulation data.
◦
Generalization performance evaluation is needed for various planetary environments and terrains.
◦
The computational cost and complexity of Bayesian deep learning models must be considered.
👍
Made with Slashpage