LiDAR Robot Navigation
LiDAR robots navigate using a combination of localization, mapping, as well as path planning. This article will introduce these concepts and explain how they work together using an example of a robot achieving its goal in a row of crops.
LiDAR sensors are relatively low power demands allowing them to extend the life of a robot's battery and decrease the amount of raw data required for localization algorithms. This allows for more iterations of SLAM without overheating GPU.
LiDAR Sensors
The core of a lidar system is its sensor which emits pulsed laser light into the surrounding. These pulses bounce off objects around them in different angles, based on their composition. The sensor measures how long it takes for each pulse to return and uses that data to calculate distances. The sensor is usually placed on a rotating platform which allows it to scan the entire area at high speeds (up to 10000 samples per second).
LiDAR sensors are classified based on the type of sensor they're designed for, whether use in the air or on the ground. Airborne lidars are often connected to helicopters or an unmanned aerial vehicles (UAV). Terrestrial LiDAR systems are typically mounted on a stationary robot platform.
To accurately measure distances the sensor must be able to determine the exact location of the robot. This information is usually gathered by a combination of inertial measurement units (IMUs), GPS, and time-keeping electronics. These sensors are used by LiDAR systems to calculate the precise location of the sensor within the space and time. This information is used to create a 3D model of the surrounding environment.
LiDAR scanners are also able to identify different surface types which is especially useful when mapping environments that have dense vegetation. When what is lidar navigation robot vacuum passes through a forest canopy it will usually generate multiple returns. Usually, the first return is associated with the top of the trees while the last return is attributed to the ground surface. If the sensor captures these pulses separately, it is called discrete-return LiDAR.
The Discrete Return scans can be used to study surface structure. For instance, a forested region could produce the sequence of 1st 2nd, and 3rd returns, with a final large pulse representing the ground. The ability to separate and record these returns in a point-cloud permits detailed models of terrain.
Once a 3D map of the environment has been created and the robot has begun to navigate using this data. This involves localization and making a path that will get to a navigation "goal." It also involves dynamic obstacle detection. This is the process of identifying obstacles that aren't present in the map originally, and adjusting the path plan in line with the new obstacles.
SLAM Algorithms
SLAM (simultaneous localization and mapping) is an algorithm that allows your robot to create an image of its surroundings and then determine the position of the robot in relation to the map. Engineers use this information for a variety of tasks, such as the planning of routes and obstacle detection.
To enable SLAM to work, your robot must have a sensor (e.g. a camera or laser), and a computer with the right software to process the data. You will also need an IMU to provide basic positioning information. The system can track the precise location of your robot in an unknown environment.
The SLAM process is a complex one and many back-end solutions are available. Whatever solution you choose, a successful SLAM system requires constant interaction between the range measurement device and the software that extracts the data, and the robot or vehicle itself. It is a dynamic process with almost infinite variability.
As the robot moves it adds scans to its map. The SLAM algorithm analyzes these scans against previous ones by using a process known as scan matching. This assists in establishing loop closures. If a loop closure is discovered it is then the SLAM algorithm makes use of this information to update its estimate of the robot's trajectory.
Another factor that makes SLAM is the fact that the environment changes as time passes. For example, if your robot is walking down an empty aisle at one point, and then comes across pallets at the next spot, it will have difficulty finding these two points on its map. This is where handling dynamics becomes crucial and is a common characteristic of modern Lidar SLAM algorithms.
SLAM systems are extremely effective at navigation and 3D scanning despite these limitations. It is particularly beneficial in situations where the robot can't rely on GNSS for its positioning for example, an indoor factory floor. It is crucial to keep in mind that even a properly configured SLAM system could be affected by mistakes. It is vital to be able to detect these errors and understand how they affect the SLAM process in order to correct them.
Mapping
The mapping function creates a map of the robot's environment. This includes the robot, its wheels, actuators and everything else that falls within its vision field. The map is used for the localization, planning of paths and obstacle detection. This is an area where 3D lidars are extremely helpful, as they can be effectively treated as an actual 3D camera (with only one scan plane).
The process of building maps may take a while, but the results pay off. The ability to build a complete and coherent map of a robot's environment allows it to move with high precision, and also over obstacles.
As a general rule of thumb, the higher resolution the sensor, the more precise the map will be. However, not all robots need high-resolution maps: for example floor sweepers may not require the same level of detail as a industrial robot that navigates factories with huge facilities.
There are a variety of mapping algorithms that can be utilized with LiDAR sensors. Cartographer is a well-known algorithm that uses a two-phase pose graph optimization technique. It corrects for drift while maintaining a consistent global map. It is especially beneficial when used in conjunction with Odometry data.
GraphSLAM is a second option that uses a set linear equations to model the constraints in diagrams. The constraints are modelled as an O matrix and an the X vector, with every vertice of the O matrix representing the distance to a landmark on the X vector. A GraphSLAM update is an array of additions and subtraction operations on these matrix elements, and the result is that all of the X and O vectors are updated to reflect new information about the robot.
Another helpful mapping algorithm is SLAM+, which combines mapping and odometry using an Extended Kalman filter (EKF). The EKF updates the uncertainty of the robot's location as well as the uncertainty of the features recorded by the sensor. This information can be utilized by the mapping function to improve its own estimation of its location and to update the map.
Obstacle Detection
A robot needs to be able to perceive its environment to avoid obstacles and reach its destination. It uses sensors such as digital cameras, infrared scans sonar, laser radar and others to detect the environment. Additionally, it employs inertial sensors to measure its speed, position and orientation. These sensors allow it to navigate in a safe manner and avoid collisions.
A range sensor is used to gauge the distance between an obstacle and a robot. The sensor can be placed on the robot, inside an automobile or on poles. It is important to keep in mind that the sensor could be affected by many factors, such as rain, wind, and fog. Therefore, it is essential to calibrate the sensor before each use.
The most important aspect of obstacle detection is identifying static obstacles. This can be accomplished by using the results of the eight-neighbor cell clustering algorithm. However this method has a low accuracy in detecting due to the occlusion caused by the distance between the different laser lines and the angular velocity of the camera which makes it difficult to recognize static obstacles in one frame. To address this issue, a technique of multi-frame fusion has been employed to improve the detection accuracy of static obstacles.
The technique of combining roadside camera-based obstacle detection with vehicle camera has proven to increase the efficiency of data processing. It also allows redundancy for other navigation operations, like planning a path. This method provides a high-quality, reliable image of the surrounding. In outdoor comparison experiments, the method was compared against other methods for detecting obstacles like YOLOv5 monocular ranging, and VIDAR.
The results of the test showed that the algorithm was able accurately identify the location and height of an obstacle, in addition to its rotation and tilt. It was also able to determine the size and color of the object. The method was also robust and steady, even when obstacles moved.