Efficient energy management optimization? How to improve the mission execution ability of drones?

Mondo Technology Updated on 2024-03-01

Drones play an important role in a variety of tasks, such as building infrastructure inspections and search and rescue in disaster environments. In these missions, collision avoidance is the key to drone motion planning.

A number of methods have been reported and verified for their effectiveness in actual drone flights. In addition, to optimize battery life, drone systems should be as light as possible.

The navigation system plays a vital role in drone control, and its process can be divided into two stages: offline and **. In the offline phase, the global path planner is an indispensable core component, which determines the flight path of the drone from the starting point to the target area.

This planner takes into account global maps and spatiotemporal wind speed information to maximize energy consumption and create a global path for energy savings. The path generation process uses the extended A* algorithm.

In the ** stage, the SLAM of LiDAR, the local path planner and the classic PD controller constitute the main components.

Among them, the SLAM algorithm based on LiDAR adopts the advanced implementation of A-LoAM, which is mainly used to provide attitude estimation and map construction of UAVs.

A-LoAM uses the point cloud data obtained by 3D LiDAR to extract feature points and generate a local map based on 3D KD tree in real time. In this process, the focus of the research is on local path planners.

Based on the waypoints generated by the global planner, the planner generates a perceptual local path. After reaching one waypoint, the drone is aimed at the next waypoint in a horizontal backward manner.

The local path planner also uses the Rapid Exploration Random Tree (RRT*) combined with the feature points scanned by the A-LOAM algorithm to quantify the perception quality and ensure the safety and stability of the flight.

New copywriting: The application of this navigation system plays an important role in drone control. Through the global path planning of the offline stage, a more efficient and energy-efficient flight path can be designed for the drone, taking into account factors such as map and wind speed.

* In the process, through the SLAM and local path planning of LiDAR, the perception and control of the UAV in the actual environment can be realized. This combination allows the drone to successfully complete complex tasks and achieve precise position control.

This study provides important technical lessons that can be further studied and improved in the future to improve the accuracy and robustness of the system.

We propose an A* algorithm diagram based on a hypersphere with radius r and an inscribed dodecahedron for path planning of UAV in 3D space.

Compared to traditional ASTS methods, our method has better performance and scalability, enabling path planning over time in high-dimensional spaces. However, the number of search nodes increases exponentially during graph construction, which makes a trade-off between the computational burden and optimal path generation.

To solve this problem, we selected 32 neighboring nodes for searching. Doing so ensures that the neighbor nodes searched from the current node are evenly distributed in all directions, allowing for better consideration of the actual effects of the wind.

Our approach is efficient, scalable, and practical, providing a reliable solution for drone path planning.

The design of this graph structure is characterized by limiting the number of adjacent nodes to achieve a nearly uniform distribution of the distance between nodes. It is worth mentioning that all 32 neighboring nodes have the same tj value because the drone is able to reach each neighboring node, ensuring the connectivity of the path.

Applying this design in the Global Planner provides an efficient and accurate strategy for drone path planning, taking into account the effects of wind. Local planners are also aware.

The local path planning module is mainly composed of the tree generation module and the path selection module of the RRT* algorithm. Firstly, a set of candidate paths was generated based on the current location and map through the RRT* algorithm.

Then, the path selection module evaluates each candidate path based on two key indicators: first, the path planning is iterated by horizon retreat until the drone reaches the designated waypoint; Second, in each planning step, the planner performs path planning within a predefined cubic area based on the map and pose estimated by A-Loam.

The RRT* algorithm is applied in the tree generation module to generate cost-optimized and collision-avoiding paths. In the study, the RRT* algorithm was further extended to generate a library of candidate paths, an approach similar to the effective exploration of drones in unknown regions.

It should be noted that if the generated tree is not smooth enough during the tracking process of the drone, then the support of dynamic model or smoothing technology is required. After the candidate path is obtained, the path selection module evaluates the feature points on each path to determine whether it can improve the accuracy of pose estimation to measure the perceived quality.

At the end node of each candidate path, the path selection module will pick out those valid feature points that can be seen by the drone, so that at the beginning node of each path planning, the scanned feature points will be defined as f.

This comprehensive evaluation and selection process helps to ensure that the drone can choose the best path in the local environment and gain better perception in actual flight. In order to consider the perceived quality of using valid feature points, the concept of a circular grid graph is introduced into the study, which allows the planner to process sparse point cloud data more efficiently, similar to how sparse point cloud interpolation (FIF) is handled.

This grid is made up of radial and angular grids, and if there is at least one valid feature point in a grid, then the grid is marked as "true".

The way to measure the quality of perception is by calculating the number of real grids approaching the drone and the distribution of those grids. The idea of the circular grid is based on two main observations: according to the research, the uncertainty of LiDAR in distance measurement will lead to large errors, so the feature points located near the UAV may be more stable for the attitude estimation of the SLAM algorithm.

According to the research, the use of feature points or constraints in different directions in SLAM algorithms can effectively prevent performance degradation. Through the perceptual quality calculation method based on circular grid map, it can provide important information for local path planning and help planners make more informed decisions, so as to achieve more accurate position estimation and higher perceptual quality during flight.

This method has been modelled and discussed and has been shown to be effective.

In order to verify the effectiveness of the proposed framework, the researchers used different implementation methods in the offline and ** processes, where the offline stage was processed with MATLAB and the ** stage was processed with a robot operating system.

By comparing it with traditional reactive path planners, this study highlights the advantages of the proposed local path planner. During the experiment, we tested three different scenarios, including photorealistic 3D models from previous studies, to fully evaluate the performance of the reactive planner and the proposer planner.

Due to the initialization process of the LiDAR SLAM module and the randomness of the RRT* algorithm, the results will be different from run to run, even with the same settings.

In order to fully evaluate the performance of each planner, two metrics were used: the simulation results showed the historical change of positional error with flight path and time, and for brevity, only three flight paths were shown.

These experimental results clearly demonstrate the superior performance of the proposed framework in different scenarios.

New Copy: This article summarizes how energy expenditure and time-of-flight perform in energy-aware paths, including the results of direct paths. The drone flies directly from the starting point to the target, and we find that the path is generated along the wind vector, resulting in a minimum of energy consumption and flight time in all three cases.

However, when the drone needs to move upwards or sideways, it cannot fly according to the wind vector. Interestingly, in the upwind data, our planner generated a path almost directly to the target area, allowing the drone to significantly reduce flight time and energy consumption.

This can be confirmed by comparing the flight time of the proposed path and the direct path. For sidewinds, the vector from the start to the target is in the opposite direction of the wind vector, so the planned path deviates from the headwind and goes around the surrounding area ( x , y ) = ( 400 , 200 ) , which leads to an increase in flight time but effectively reduces energy consumption.

Long-range air shows.

After comparing the results of global planning and gazebo simulations, we found that the actual flight time was about 1000 seconds less than the expected flight time. To solve this problem, we have adjusted the trade-off factor a.

Especially for long-range flight scenarios, the adjustment of the weight factor A also applies. We find that in each mission between two waypoints, the value of a is in the range [0.].1, 0.4, 0.7] within the variation.

As can be seen in Figure 19, there is a clear relationship between a and the energy expenditure at each waypoint. Between waypoints 1 and 2, the perception planner's = 07 makes the UAV maintain the perceived quality, resulting in an increase in energy consumption.

But when a is 01, energy consumption is minimized, which means that adjusting a affects not only the accuracy of attitude estimation, but also the energy consumption of the drone. Between waypoints 2 and 3, the planner's = 01 Causes flight failure due to incorrect positioning.

Between waypoints 3 and 4, the planner's = 04 Reduced energy consumption by 27% compared to before, but caused flight failure. In order to circumvent this, we need to set a according to the environment around the waypoint.

In landmark-rich areas, drones can effectively save energy consumption by lowering A. However, in featureless regions, we suggest adding a to maintain a high perceived quality.

By adjusting this sensitive parameter, our framework can map out local pathways for energy savings while maintaining high-quality perception.

Related Pages