Article Directory
Study Notes-Introduction and Architecture of SLAM
This note is based on the video study summary of station B for everyone to learn, exchange and discuss.
1. The proposal and development of SLAM
1. In order to solve two problems: the location of the current environment and the state of the current environment.
2. In the early days, some scholars proposed the use of probability estimation methods to solve the problem of positioning and composition.
3. Some parts of SLAM: visible light vision, IMU, lidar, sonar
4. Some applications of SLAM: unmanned driving, VR, AR, UAVs, robots
classification 5.SLAM of: EKF-SLAM, FAST-SLAM , Graph-SLAM , etc.
Second, talk about SLAM from the filter
1. Make positioning based on probability estimation
1. Estimate the position based on its own motion model, and express the variance through the Gaussian model
2. The sensor obtains position estimation based on the map
3. Realize fusion through a priori + a posteriori
4. Give the prediction model and observation model of the robot position:
2. Continuously update its position through probability
1. The probability at the starting time is the average value
. 2. Through observation, the probability near the location with the road sign is:
update the state at this time:
3. Continue to update the location:
4. Obtain the observation data of the road sign again and update the status:
3. A new breakthrough in SLAM-graph optimization
1.
2.
3.
Fourth, the knowledge structure of SLAM
1. Overall framework
2. Sensor
1. Classification of sensors: vision, laser, sonar ultrasound, inertial navigation, etc.
3. Basic theory
1. Filtering (random estimation)
2. Coordinate system, rigid body motion,
Euclidean rotation and quaternion. For details, see "Fourteen Lectures on Visual SLAM"
3. Camera model, visual geometry
Camera model:
visual geometry:
4. Pose estimation and fusion
PnP ICP RANSAC…
5. Loop and graph optimization
1. Bag of words model:
6. Environmental expression and map construction
Point cloud map, topological map, semantic map, feature map...
7. Deep learning
Five, a brief introduction to ROS
1. Structure function: process management, internal process communication, module drive
2. Tools: simulation, display, image interaction, data recording
3. Realizable functions: control, planning, perception, composition, operation
4. Resources: open source basic development kit , Updated version, open source tutorial