Outline
?
?
S?ren: SLAM
– SLAM introduction
–SIFT SLAM
– Experimental Results
Vikash: SIFT matching
Vision-based SLAM
Mobile Robot Localization And Mapping With
Uncertainty using Scale-Invariant Visual Landmarks
- Se, Lowe, Little
Vikash Mansinghka & S?ren Riisgaard
?
?
?
–
–
?
?
Simultaneous Localization And Mapping
The SLAM problem
Preconditions
Enough landmarks
Static landmarks
State Est. vs SLAM
State Estimation
– EKF
– HMM – Viterbi
– HMM – Particle filters
?SLAM
– Map and robot pose is coupled
– Errors are correlated
3 SLAM Algorithms
? EKF based SLAM
? FastSLAM
?SIFT SLAM
? Comparison
EKF FastSLAM SIFT SLAM
Robot Pose EKF Particle Filter Least Squares EKF
Landmarks Combined with pose 1 Kalman Filter per
Landmark/sample
1 Kalman Filter per
Landmark
Performance O(K
2
) O(M K) / O(M log K) O(K) ?
Applications Small scenarios Large Scenarios Vision
Observation Landmarks Landmarks Robot pose
K = Landmarks, M = Particles
SIFT SLAM
?
–
?
–
?
–
?
– l
Odometry based state estimate
Where did I try to go?
Least Squares localization estimate
Where did I go?
Localization – EKF
Where did I really go?
Mapping
Update andmark cov, add new landmarks