abstract |
Method and system for predicting trajectories of agents n A computer-implemented recurrent neural network system (10) for predicting trajectories of agents present in a vicinity of a vehicle comprising traffic lanes, the system comprising: n- a first encoder module (MOD_ENC_1) configured to determine, for each agent and for a first time step, a vector (h_{0,k}) at least representative of a social interaction influence that the other agents have on said agent, n- a second encoder module (MOD_ENC_2) configured to determine a matrix (M_L) representative of similarities between the traffic lanes, n- a decoder module (MOD_DEC) comprising configure to determine, for each agent and for said first time step, at least a positional representation (x_{1,k}) of said agent. |