LiDAR Robot Navigation
LiDAR robots navigate using a combination of localization, mapping, and also path planning. This article will outline the concepts and show how they work using an example in which the robot reaches an objective within a row of plants.
LiDAR sensors are relatively low power requirements, allowing them to increase a robot's battery life and decrease the amount of raw data required for localization algorithms. This allows for a greater number of variations of the SLAM algorithm without overheating the GPU.
LiDAR Sensors
The central component of lidar systems is its sensor, which emits laser light pulses into the environment. These light pulses strike objects and bounce back to the sensor at a variety of angles, based on the structure of the object. The sensor monitors the time it takes each pulse to return, and utilizes that information to determine distances. Sensors are placed on rotating platforms, which allow them to scan the surroundings quickly and at high speeds (10000 samples per second).
LiDAR sensors can be classified according to whether they're designed for airborne application or terrestrial application. Airborne lidar systems are typically mounted on aircrafts, helicopters, or unmanned aerial vehicles (UAVs). Terrestrial LiDAR is usually mounted on a stationary robot platform.
To accurately measure distances, the sensor must always know the exact location of the robot. This information is gathered by a combination inertial measurement unit (IMU), GPS and time-keeping electronic. These sensors are utilized by LiDAR systems in order to determine the precise location of the sensor within the space and time. This information is used to create a 3D model of the surrounding.
LiDAR scanners are also able to recognize different types of surfaces and types of surfaces, which is particularly useful for mapping environments with dense vegetation. When a pulse passes a forest canopy, it will typically register multiple returns. Typically, the first return is attributable to the top of the trees, and the last one is attributed to the ground surface. If the sensor records these pulses in a separate way, it is called discrete-return LiDAR.
Discrete return scanning can also be helpful in studying surface structure. For instance, a forested region might yield the sequence of 1st 2nd and 3rd returns with a final large pulse that represents the ground. The ability to separate these returns and record them as a point cloud allows for the creation of detailed terrain models.
Once a 3D model of the surroundings has been created and the robot has begun to navigate using this information. This involves localization, building an appropriate path to reach a goal for navigation,' and dynamic obstacle detection. The latter is the process of identifying new obstacles that aren't present on the original map and then updating the plan accordingly.
SLAM Algorithms
SLAM (simultaneous mapping and localization) is an algorithm which allows your robot to map its surroundings, and then identify its location relative to that map. Engineers use this information for a variety of tasks, such as the planning of routes and obstacle detection.
To enable SLAM to function, your robot must have sensors (e.g. a camera or laser) and a computer with the right software to process the data. Also, you will require an IMU to provide basic information about your position. The result is a system that will accurately determine the location of your robot in an unknown environment.
The SLAM process is extremely complex and many back-end solutions exist. Regardless of which solution you select for your SLAM system, a successful SLAM system requires constant interaction between the range measurement device and the software that collects the data, and the robot or vehicle itself. This is a highly dynamic process that has an almost infinite amount of variability.
When the robot moves, it adds scans to its map. The SLAM algorithm compares these scans to prior ones making use of a process known as scan matching. This allows loop closures to be identified. When a loop closure has been detected, the SLAM algorithm uses this information to update its estimated robot trajectory.
The fact that the surroundings can change over time is a further factor that complicates SLAM. For instance, if your robot travels through an empty aisle at one point and then encounters stacks of pallets at the next location, it will have difficulty matching these two points in its map. This is when handling dynamics becomes important and is a typical characteristic of the modern Lidar SLAM algorithms.
Despite these difficulties however, a properly designed SLAM system is incredibly effective for navigation and 3D scanning. It is especially beneficial in situations where the robot isn't able to depend on GNSS to determine its position, such as an indoor factory floor. It is important to note that even a properly configured SLAM system can experience errors. It is essential to be able to detect these errors and understand how they impact the SLAM process in order to correct them.
Mapping
The mapping function creates a map of a robot's surroundings. This includes the robot and its wheels, actuators, and everything else within its vision field. The map is used to perform the localization, planning of paths and obstacle detection. This is a domain in which 3D Lidars are particularly useful as they can be regarded as an 3D Camera (with one scanning plane).
The process of building maps takes a bit of time however the results pay off. The ability to create a complete, consistent map of the robot's environment allows it to perform high-precision navigation, as well as navigate around obstacles.
In general, the higher the resolution of the sensor, then the more accurate will be the map. Not all robots require maps with high resolution. For instance floor sweepers may not require the same level detail as a robotic system for industrial use navigating large factories.
This is why there are a number of different mapping algorithms to use with LiDAR sensors. Cartographer is a popular algorithm that utilizes a two-phase pose graph optimization technique. It corrects for drift while maintaining an unchanging global map. It is especially useful when paired with odometry data.
robot vacuum with lidar and camera is GraphSLAM that employs a system of linear equations to represent the constraints of a graph. The constraints are modelled as an O matrix and an the X vector, with every vertex of the O matrix representing the distance to a landmark on the X vector. A GraphSLAM Update is a series subtractions and additions to these matrix elements. The result is that both the O and X vectors are updated to account for the new observations made by the robot.
SLAM+ is another useful mapping algorithm that combines odometry with mapping using an Extended Kalman filter (EKF). The EKF updates the uncertainty of the robot's location as well as the uncertainty of the features mapped by the sensor. The mapping function will utilize this information to better estimate its own position, which allows it to update the base map.
Obstacle Detection
A robot needs to be able to perceive its environment to avoid obstacles and get to its destination. It uses sensors like digital cameras, infrared scanners, laser radar and sonar to determine its surroundings. It also uses inertial sensor to measure its speed, location and its orientation. These sensors aid in navigation in a safe manner and avoid collisions.
A key element of this process is obstacle detection, which involves the use of a range sensor to determine the distance between the robot and the obstacles. The sensor can be attached to the vehicle, the robot or a pole. It is important to keep in mind that the sensor could be affected by a variety of elements, including rain, wind, and fog. Therefore, it is important to calibrate the sensor before each use.
An important step in obstacle detection is identifying static obstacles, which can be done by using the results of the eight-neighbor cell clustering algorithm. This method isn't very precise due to the occlusion created by the distance between the laser lines and the camera's angular velocity. To solve this issue, a method of multi-frame fusion was developed to increase the accuracy of detection of static obstacles.
The method of combining roadside unit-based as well as obstacle detection using a vehicle camera has been proven to improve the efficiency of processing data and reserve redundancy for subsequent navigation operations, such as path planning. The result of this technique is a high-quality image of the surrounding area that is more reliable than a single frame. The method has been tested against other obstacle detection methods, such as YOLOv5, VIDAR, and monocular ranging, in outdoor comparative tests.
The results of the experiment showed that the algorithm was able correctly identify the position and height of an obstacle, in addition to its tilt and rotation. It was also able identify the size and color of an object. The method also showed excellent stability and durability, even when faced with moving obstacles.