---
title: Obstacle Detection using ML
---
### Big idea

We have a depth camera, so if we can detect obstacles within an image (using bounding boxes or pixelwise segmentation), we can determine the **pose and dimensions** of the obstacles that can be fed into the map (occupancy grid).

OR

Generalize to update the occupancy grid for any obstacle in a certain radius in front of the vehicle.

### Potential Approaches
- A combined method of crater detection and recognition based on deep learning (https://www.tandfonline.com/doi/full/10.1080/21642583.2020.1852980)
- Hough transform (feature extraction technique used in image analysis, computer vision, and digital image processing)
- radial consistency method
- Cerbereus SubT Challenge Papers (http://www.subt-cerberus.org/publications.html)

- NERF - Neural Radiance Fields, another way to represent the environment using just vision (https://mikh3x4.github.io/nerf-navigation/)

- JARTool - (paper) https://link.springer.com/content/pdf/10.1023/A:1007400206189.pdf
- Crater Detection Paper https://www.sciencedirect.com/science/article/pii/S0273117707003675
- Domain randomization
- data augmentation (http://proceedings.mlr.press/v119/jun20a/jun20a.pdf)

### Building Datasets
- https://mars.nasa.gov/news/8689/nasas-mars-rover-drivers-need-your-help/

### Cool Links
- https://evjang.com/2021/10/23/generalization.html