Authors: Mark Maimone, Yang Cheng, and Larry Matthies
Year: 2007
- 6DOF VO by tracking features between stereo images
- Impossible to use VO while driving "fast"
- Helps to estimate slip
- Tracks Harris features with the highest score and compute their 3D position with stereo vision
- Features tracked with the 3D position and odometers, perform a corelation based search for matching, remove outliers with ransac and a first motion estimation is performed
- Then the robot motion is refined with ML estimation
- cam : 640x480px 80x64° and 8.4cm baseline
- "update often, search less" doesn't work and other operationnal constraints
- IMU drifted slowly + good encoder based estimation
- VO used in high sloped : more features and risk to slip
- VO increased precision and safety of the navigation
- Navigating with rover track: slip check by looking backward
- Slip checking device
- Need to implement the search for a textured enough terrain
- Necessity of wide angles camera for better feature tracking
Bilan de deux ans de mission sur Mars: de belles anecdotes et une preuve de la nécessité de la navigation visuelle. Element de comparaison intéressant.