Robot trajectory prediction method based on road topology connection relation

The invention belongs to the technical field of robot trajectory prediction. The invention provides a robot trajectory prediction method based on a road topology connection relation. The method comprises the steps of obtaining road information, a historical track of a to-be-predicted robot and historical tracks of surrounding robots through sensing of related sensors, encoding and then decoding the road information, the historical track of the to-be-predicted robot and the historical tracks of the surrounding robots by using a recurrent neural network to obtain a predicted track and a corresponding probability, and the neural network is trained by designing a loss function to obtain parameters of the prediction model. According to the embodiment of the invention, influence factors such as road information of a surrounding map of the to-be-predicted robot, interaction between the to-be-predicted machine and the surrounding robot, information of the to-be-predicted machine and the like can be considered, the pr