Robot Navigation in Human-Robot Shared Environments Based on Social Interaction Model

Robot navigation in a dynamic environment shared with humans is a challenging task. Robot needs to process surrounding environmental data in real time and respond quickly. However, as the number of humans increases, the robot’s sensors recognize more obstacles, and the amount of information that the navigation algorithm needs to process also increases, resulting in a decrease in the performance of robot navigation. To solve this problem, we propose a robot navigation algorithm for crowd environments. This algorithm employs a model-based deep reinforcement learning approach as the underlying framework and utilizes graph convolutional networks to encode high-order interactions between people and robot as the state input for the deep reinforcement learning algorithm. We integrate multi-head attention mechanism into the graph convolutional network to assign weights to graph network nodes that are more in line with social behavior norms, and we also introduce a gating mechanism into the model to make it more lightweight. The incorporation of these mechanisms can significantly enhance the navigation performance of robot in dense crowds. We evaluated our method based on advanced crowd navigation baseline algorithms and confirmed its effectiveness.