Multi-agent communication based on adaptive discretization

In real-world application scenarios, communication in multi-agent systems faces numerous challenges, especially issues related to noise interference. Taking autonomous driving systems as an example, when multiple autonomous vehicles interact and communicate with each other in the same environment, they inevitably encounter interference from external environments and noise from internal systems. These interferences can originate from complex road conditions, electromagnetic interference from surrounding electronic devices, or even signal attenuation caused by changes in weather. These factors can negatively affect the quality of communication between vehicles, thereby impacting the decision-making and coordination ability of the autonomous driving systems and increasing the risk of accidents. To address the issue of noise interference in real data, this paper proposes an adaptive feature fusion communication algorithm based on the VQVAE architecture. In this work, data is first passed through an encoder to obtain the original continuous vector. Then, this continuous vector is processed through a linear network layer to learn the parameter tensor hα. Subsequently, the obtained original continuous vector is discretized to get a discretized result. Next, the learned hα vector is multiplied by the discretized result. The purpose of this operation is to control the degree of discretization based on the learned hα parameter. At the same time, we calculate the value of (1-hα) and then multiply it with the original continuous vector to retain a part of its continuity. Finally, these two parts are added together and sent to the receiving agent (i.e., decoder). Compared with manually setting the degree of discretization and the original AE, VQVAE models, experimental results on three public datasets demonstrate that the algorithm proposed in this study can effectively improve the performance of multi-agent communication.