A 3D Crowd Exploration Method Based on Multi-Head Attention Asynchronous Reinforcement Learning

Abstract

The invention discloses a three-dimensional group exploration method based on multi-head attention asynchronous reinforcement learning. It includes the following steps: Step 1. The main process of the command center sets up a shared sample multiplexing cache and initializes the benchmark exploration strategy; Step 2. The command center starts the sub-process; Step 3. The command center uses pixel control algorithm to optimize the unmanned data based on the shared sample multiplexing cache. Machine exploration strategy; step 4, the command center uses the trust domain strategy algorithm to obtain the flight trajectory of the drone group based on the shared sample multiplexing cache; step 5, repeatedly execute steps 2, 3, and 4 until the trajectory of the drone group does not change; Step 6. The command center sends the optimal trajectory mobilization instruction to the UAV group. The invention solves the problem of low sample sampling efficiency of the reinforcement learning algorithm, and the algorithm achieves a better data collection effect when using the same number of samples for learning, and further obtains an optimal trajectory for maximizing data collection.

Description

technical field [0001] The invention belongs to the field of mobile group perception, and in particular relates to a three-dimensional group exploration method based on multi-head attention asynchronous reinforcement learning. Background technique [0002] Mobile group sensing technology is currently developing rapidly and supports the data acquisition needs of smart cities. Mobile group sensing technology uses mobile devices used by a large number of users as the basic sensing unit, and cooperates through the mobile Internet to form an interactive and participatory sensing network, realize sensing task distribution and data collection and utilization, and finally complete large-scale and complex Social sensing tasks to help professionals or the public collect data, analyze data, and share data. However, the mobile group sensing system based on mobile devices is often affected by many aspects, such as the uncertainty of user movement and the quality of mobile devices. These...

AI Technical Summary

Problems solved by technology

[0004] 1. It is difficult to model complex application environments. The actual application scenarios of mobile group sensing are often dynamic and complex. For example, mobile group sensing data collection for post-disaster rescue. According to the results of environmental modeling, the current UAV swarm flight trajectory is reasonably planned for data collection tasks. Therefore, the accuracy of environmental modeling greatly affects the completion quality of group perception tasks. How to accurately and quickly create space for real application environments Die has become a big problem;

[0005] 2. Insufficient exploration of three-dimensional space. Aiming at the insufficient exploration caused by the explosion of three-dimensional space, it is necessary to design a reasonable, stable and efficient exploration mechanism to promote the drone group to quickly and efficiently perceive the entire unknown three-dimensional mobile group perception scene. Exploration to improve the quality and efficiency of drone swarm environment modeling and optimal trajectory search efforts

[0006] 3. The utilization rate of reinforcement learning samples is low. Existing reinforcement learning algorithms are faced with the problem of extremely low sample utilization rate, and cannot effectively and fully learn from the only samples. In reality, the cost of sample sources for 3D mobile group perception tasks is high , The acquisition speed is slow, how to make the algorithm more effective and fully sample and learn the existing samples without affecting the learning effect of the algorithm is an urgent problem to be solved

Images