LiDAR Robot Navigation

LiDAR robots navigate using a combination of localization and mapping, as well as path planning. This article will introduce these concepts and explain how they work together using an easy example of the robot achieving a goal within a row of crops.

LiDAR sensors are low-power devices that prolong the life of batteries on robots and reduce the amount of raw data needed for localization algorithms. This allows for a greater number of iterations of the SLAM algorithm without overheating the GPU.

LiDAR Sensors

The sensor is the core of the Lidar system. It releases laser pulses into the environment. These light pulses strike objects and bounce back to the sensor at a variety of angles, depending on the structure of the object. The sensor determines how long it takes each pulse to return, and uses that data to calculate distances. The sensor is typically placed on a rotating platform, permitting it to scan the entire area at high speeds (up to 10000 samples per second).

LiDAR sensors can be classified based on the type of sensor they're designed for, whether use in the air or on the ground. Airborne lidars are often attached to helicopters or unmanned aerial vehicles (UAV). Terrestrial LiDAR is usually installed 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 captured by a combination of an inertial measurement unit (IMU), GPS and time-keeping electronic. These sensors are used by LiDAR systems to determine the precise location of the sensor within the space and time. The information gathered is used to create a 3D representation of the surrounding.

LiDAR scanners can also identify different types of surfaces, which is especially useful when mapping environments with dense vegetation. When a pulse passes through a forest canopy, it is likely to generate multiple returns. The first one is typically attributed to the tops of the trees, while the second is associated with the ground's surface. If the sensor records each pulse as distinct, this is called discrete return LiDAR.

Distinte return scanning can be helpful in analyzing the structure of surfaces. For instance forests can produce a series of 1st and 2nd returns, with the final large pulse representing the ground. The ability to separate and record these returns in a point-cloud permits detailed terrain models.

Once a 3D model of environment is constructed, the robot will be able to use this data to navigate. This process involves localization and building a path that will reach a navigation "goal." It also involves dynamic obstacle detection. This is the process of identifying obstacles that aren't visible in the original map, and updating the path plan accordingly.

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 the data for a variety of purposes, including planning a path and identifying obstacles.

To utilize SLAM your robot has to have a sensor that gives range data (e.g. A computer with the appropriate software to process the data, as well as either a camera or laser are required. Also, you need an inertial measurement unit (IMU) to provide basic positional information. The result is a system that can precisely track the position of your robot in an unknown environment.

The SLAM process is a complex one and many back-end solutions exist. Whatever solution you choose for a successful SLAM, it requires constant communication between the range measurement device and the software that extracts the data and the vehicle or robot. This is a highly dynamic process that can have an almost infinite amount of variability.

When the robot moves, it adds scans to its map. The SLAM algorithm analyzes these scans against previous ones by using a process called 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.

Another factor that complicates SLAM is the fact that the environment changes in time. For instance, if your robot is walking down an aisle that is empty at one point, and it comes across a stack of pallets at a different location, it may have difficulty finding the two points on its map. This is where the handling of dynamics becomes critical and is a standard characteristic of modern Lidar SLAM algorithms.

SLAM systems are extremely efficient at navigation and 3D scanning despite these limitations. It is particularly useful in environments that do not allow the robot to rely on GNSS-based positioning, like an indoor factory floor. However, it is important to remember that even a well-designed SLAM system may have mistakes. It is vital to be able recognize these flaws and understand how they affect the SLAM process in order to rectify them.

Mapping

The mapping function builds an image of the robot vacuum with lidar and camera's surrounding which includes the robot itself including its wheels and actuators as well as everything else within its view. The map is used for the localization, planning of paths and obstacle detection. This is an area in which 3D lidars are particularly helpful because they can be utilized as the equivalent of a 3D camera (with one scan plane).

The process of building maps can take some time however the results pay off. The ability to build an accurate and complete map of the robot's surroundings allows it to navigate with high precision, as well as around obstacles.

As a general rule of thumb, the greater resolution the sensor, more accurate the map will be. Not all robots require maps with high resolution. For example, a floor sweeping robot may not require the same level of detail as an industrial robotics system operating in large factories.

To this end, there are a number of different mapping algorithms for use with LiDAR sensors. Cartographer is a very popular algorithm that employs a two phase pose graph optimization technique. It corrects for drift while maintaining a consistent global map. It is especially useful when paired with odometry.

GraphSLAM is a different option, that uses a set linear equations to represent the constraints in a diagram. The constraints are modelled as an O matrix and a X vector, with each vertex of the O matrix representing the distance to a landmark on the X vector. A GraphSLAM Update is a sequence of subtractions and additions to these matrix elements. The result is that all the O and X Vectors are updated to take into account the latest observations made by the robot.

SLAM+ is another useful mapping algorithm that combines odometry and mapping using an Extended Kalman filter (EKF). The EKF updates not only the uncertainty in the robot's current position but also the uncertainty in the features drawn by the sensor. The mapping function can then utilize this information to estimate its own position, which allows it to update the underlying map.

Obstacle Detection

A robot must be able to perceive its surroundings in order to avoid obstacles and get to its desired point. It makes use of sensors such as digital cameras, infrared scanners, LiDAR robot navigation laser radar and sonar to determine its surroundings. It also makes use of an inertial sensors to determine its position, LiDAR Robot Navigation speed and its orientation. These sensors help it navigate in a safe manner and prevent collisions.

One of the most important aspects of this process is the detection of obstacles, which involves the use of a range sensor to determine the distance between the robot and the obstacles. The sensor can be positioned on the robot, in the vehicle, or on poles. It is important to remember that the sensor is affected by a myriad of factors, including wind, rain and fog. It is essential to calibrate the sensors prior each use.

The most important aspect of obstacle detection is the identification of static obstacles. This can be done by using the results of the eight-neighbor cell clustering algorithm. This method isn't very precise due to the occlusion induced by the distance between laser lines and the camera's angular velocity. To address this issue, a method of multi-frame fusion has been used to increase the accuracy of detection of static obstacles.

The method of combining roadside unit-based as well as obstacle detection by a vehicle camera has been shown to improve the efficiency of processing data and reserve redundancy for further navigational operations, like 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 compared against other obstacle detection methods, such as YOLOv5, VIDAR, and monocular ranging in outdoor tests of comparison.

The results of the experiment proved that the algorithm could correctly identify the height and location of obstacles as well as its tilt and rotation. It was also able identify the color and size of the object. The method also exhibited solid stability and reliability even in the presence of moving obstacles.