Abstract: In the coordinated control of swarm robots using swarm intelligence methods for target search, the search behavior of the robots is guided by both their own perception of the environment and the shared experience within the group, with group experience playing a significant role. Since group experience essentially belongs to the cognition of an individual robot, this paper introduces a collective decision—the estimation of the target's position—under the absolute localization mechanism to improve target search speed. First, an extended particle swarm optimization (PSO) model is used to model the swarm robot system. Then, based on the nature of the guiding role of group experience in search behavior and the essence of using RSSI (Real-Side Scale Indicator) to estimate the target position, the experience, which is essentially that of a single robot, is combined with collective decision-making for target search control. When the target position estimation condition is met, the estimated target position is introduced into the extended PSO model; otherwise, the original model is retained. Simulation results show that compared with the model method that simply uses group experience, the proposed algorithm outperforms the former in terms of target search success rate, search efficiency, and energy consumption.
Keywords: swarm robots; particle swarm optimization algorithm; target search; position estimation
Targetpositionestimationaidedswarmroboticsearchunderconditionsofabsolutelocalizationmechanism
1 Introduction
Swarm robots are special multi-autonomous robot systems [1], composed of several relatively simple robots, each with limited environmental perception capabilities, and their structure and functional roles are isomorphic. Swarm robot systems have the following characteristics [2]: robustness; flexibility; scalability of system size. Swarm robot research is divided into several benchmark problems [3], such as transportation, formation, search, and encirclement. Among them, the search task includes several essential scientific problems [4], such as localization, communication, collision avoidance, and path planning. Localization is the basis for robots to identify their own position in the environment and work collaboratively with other robots. Swarm robot localization [5] refers to the process by which autonomous mobile robots collect measurement data from odometers and sensors in real time during movement, and infer their own pose and the pose of robots in the swarm with time-varying characteristics. On this basis, the running speed can also be inferred and the target pose can be estimated. Swarm robot localization is mainly divided into two types [6]: absolute localization and relative localization. Absolute positioning of swarm robots [7] involves setting a reference point in the working environment or outside the environment as the reference benchmark for all robots. Each robot then calculates its pose using its own sensors and corrects the calculated position using the reference point to eliminate accumulated errors. The relative positioning mechanism of swarm robots [8] involves each individual robot using its own position as the reference point and its own head orientation as the positive direction of the coordinate system to construct its own local coordinate system. The relative position detection of the robot with respect to other robots is used as the pose of other robots. Absolute positioning methods mainly include navigation beacon positioning, active or passive marker positioning, image matching positioning, GPS positioning, and probabilistic positioning. Relative positioning mainly includes inertial navigation and range measurement methods [9]. This paper studies the target search problem of swarm robots under the absolute positioning mechanism. Existing research on target search of swarm robots has adopted an extended particle swarm algorithm model for system modeling and control. The extended particle swarm model relies on its own experience and group experience to iterate the expected position and velocity. Individual experience is determined by introducing a short-term memory mechanism, based on the target signal strength at the current and previous positions. Group experience, on the other hand, is determined by the target signal strength values monitored by the robot and other robots in its time-varying group. Thus, group experience is derived from the cognitive "election" of all robots in the time-varying group. Essentially, both individual robot perception and group experience guide the robot's search behavior through the cognition of a single robot. Wireless sensor networks consist of distributed nodes, each with sensing, computing, and communication capabilities, functionally similar to swarm robots. Node localization techniques in wireless sensor networks mainly include ranging-free and ranging-required methods, with the RSSI algorithm in ranging-required localization being a relatively mature approach. Swarm robots are essentially wireless sensor networks, with individual robots being nodes in these networks possessing motion attributes. Target localization based on wireless sensor networks is essentially the result of group decision-making. Considering search efficiency, this paper combines target position estimation with swarm robot search. When the target position estimation conditions of the RSSI method are met, the target position is estimated by detecting the target signal through the robot's sensors, and the target position estimate is used to correct the model, thereby assisting the swarm robots in target search.
This paper is structured as follows: Section 2 describes a method for target search in swarm robots using an extended particle swarm optimization (PSE) algorithm for system modeling and coordinated control, and analyzes the essence of guiding individual robot search behavior based on this method; Section 3 introduces the RSSI method and analyzes its localization essence in conjunction with autonomous motion robots; Section 4 elaborates on the swarm robot search method based on target position estimation and presents its algorithm; Section 5 presents simulation experiments and discusses the results. For comparison, a target search experiment using an extended particle swarm optimization model that does not consider target position estimation was also conducted. Section 6 draws conclusions and concludes with a prospect for future research.
2. Modeling and Coordinated Control of Swarm Robots for Target Search
The target search task of swarm robots occupies a particularly important position in swarm robot applications. Compared with other tasks, target search is fundamental. In the coordinated control of the target search task of swarm robots, the extended particle swarm model is used as its control tool. The extended particle swarm model [10] is used as the target search task of swarm robots.
At the beginning of the target search task, the robot is randomly placed in a circle with a radius smaller than R, and the center of the circle is far from the target position, in order to increase the search difficulty. The robot's initial speed and position are both random values, and the initial speed is a random value between [0,1] and the maximum speed. The robot monitors the target signal for the first time, and the target signal strength value [11] is determined by...
It is normally distributed Gaussian white noise. The robot broadcasts the measured signal strength value and its own position coordinates, while simultaneously listening for signals sent by other robots in the swarm, which exhibit time-varying characteristics. The robot takes its initial position as its optimal position, compares the target signal strength value with that of its neighboring robots in the swarm, and takes the position of the robot with the highest value as its optimal position in the swarm. Based on the monitored signals, if no robot in the swarm detects the target signal, it enters a spiral divergent roaming state, with the iterative formula being:
If at least one robot in the group detects the target strength signal, it enters an intelligent search state. Its iterative formula is Equation 1. The robot calculates its desired position and speed based on its initial position, its own optimal position, and the group's optimal position. At this point, the robot moves one step. The robot monitors the target signal again and calculates the target signal strength value according to Equation 2. The robot broadcasts the measured signal strength value and updated position coordinates, while simultaneously listening to the signals sent by other robots in the group that exhibit time-varying characteristics. The robot uses [12] to...
The robot calculates its optimal position and compares its target signal strength value with that of neighboring robots in the group. The position of the robot with the highest value is taken as its optimal group position. Based on the monitored signals, if no robot in the group detects the target signal, it enters a spiral divergent roaming state, with the iterative formula being Equation 3. If at least one robot in the group detects the target signal strength, it enters an intelligent search state, with the iterative formula being Equation 1. The robot calculates its desired position and speed based on its current position, its own optimal position, and the group's optimal position. At this point, the robot moves another step. This process is repeated until at least one robot in the group is less than a set value away from the target, or the maximum number of iterations has been exceeded.
Individual experience is determined by introducing a short-term memory mechanism, based on the target signal strength at the current and previous positions. Group experience, on the other hand, is determined by the target signal strength values monitored by the robot and other robots in its time-varying group. Thus, group experience is attributed to the cognition of one robot within the time-varying group. Essentially, both individual robot perception and group experience guide the robot's search behavior through the cognition of a single robot, without collective decision-making. Target position estimation, however, is equivalent to collective decision-making because it utilizes the experience of at least three robots involved in the estimation.
3. Target search with target location estimation
Wireless sensor networks consist of distributed nodes, each with sensing, computing, and communication capabilities, functionally similar to swarm robots. Node localization techniques in wireless sensor networks mainly fall into two categories: ranging-free and ranging-dependent. The RSSI algorithm in ranging-dependent localization is a relatively mature method. Based on the similarity between wireless sensor network node localization techniques and swarm robot absolute localization techniques, the RSSI algorithm is introduced. When a robot is searching for a target, the RSSI condition for estimating the target position is met if at least three robots in the swarm (including itself) detect a target signal strength that is not zero, and these three robots are not on the same straight line.
4. Algorithm Description
The simulation algorithm process for the new model introducing target position estimation is as follows:
5 Simulation Results
For both the extended model and the proposed model, fifty simulations were conducted using 3 to 10 robots each, totaling 8000 simulations. The simulation data were then analyzed and compared. The performance evaluation metrics for the extended particle swarm model and the proposed model mainly include: a comparison of the target search success rate of the proposed model with 3 to 10 robots; a comparison of the expected number of steps required for the target search using the proposed model with 3 to 10 robots; a comparison of the target search success rate of the proposed model and the extended particle swarm model with 3 to 10 robots; a comparison of the expected number of steps required for the target search using the proposed model and the extended particle swarm model with 3 to 10 robots; a comparison of the expected step length of the target search using the proposed model with 3 to 10 robots; and a comparison of the estimated target position of an individual robot and the expected distance from the target to the optimal swarm position when the target position is estimated using the RSSI method.
5.2 Simulation Data
For the sake of convenience in comparing the model in this paper with the extended particle swarm model, and to maintain a neat style in the data graphs and tables, the model in this paper will be referred to as Method 1, and the extended particle swarm model will be referred to as Method 2.
As can be seen from the table and graph, the success rate of Method 1 increases with the number of robots, indicating that the efficiency of swarm search improves as the system scales up. With the same number of robots, Method 1 is generally more efficient than Method 2, demonstrating that Method 1 has higher search efficiency than Method 2 under the same conditions.
As can be seen from the table and graph, the number of search steps for Method 1 gradually decreases as the number of robots increases, indicating that the number of steps required for group search is decreasing and its search efficiency is increasing. When the number of robots is the same, Method 1 consumes slightly fewer steps than Method 2, but its standard deviation is significantly better than Method 2, indicating that Method 1 has higher stability in terms of search steps than Method 2.
As can be seen from the table and figure, when the number of robots is the same, method 1 consumes fewer steps than method 2, indicating that it consumes less energy and demonstrating its advantage in energy saving.
When swarm robots search for a target using Method 1, they estimate the target position when the RSSI (Real-Side Scale Indicator) condition is met, and then update the expected position and velocity accordingly. If the RSSI condition is not met, the iterative formula in Method 2 is used. The ability of the new iterative formula in Method 1 to guide the swarm robots in target searching can be judged by the expected distance between the estimated target position and the swarm's optimal position when the RSSI condition is met. Data from the table and graph shows that, with the same number of robots, Method 1, when satisfying the RSSI condition for estimating the target position, provides a smaller estimated distance from the target than the distance between the swarm's optimal position and the target, indicating a stronger ability to guide the robots in target searching.
6. Conclusion
Simulation experiments verified the feasibility of the proposed model, which outperformed the Extended Particle Swarm Optimization (EPSO) model in terms of localization efficiency, number of steps, and energy consumption. However, this study only examines the localization efficiency of target search tasks under the absolute localization mechanism; further research is needed on the localization efficiency of introducing target position estimation under the relative localization mechanism.
Name: Zan Yunlong Address: Room 405, Graduate Student Apartment, No. 66 Walu Road, Taiyuan City, Shanxi Province Postcode: 030024
Department: Graduate School of Systems Engineering, Taiyuan University of Science and Technology Tel: 15513079361 Email: [email protected]
