Trajectory generation for mobile agents

A computer system for generating a trajectory for a first agent of a plurality of agents navigating a mapped area, the computer system comprising:

• one or more processors;
• and
• memory coupled to the one or more processors, the memory embodying computer-readable instructions, which, when executed on the one or more processors, cause the one or more processors to:
• receive an observed state of each of the plurality of agents, and map data of the mapped area;
• generate an initial estimated trajectory for each of the plurality of agents based on the observed state of each agent and the map data;
• perform a first collision assessment to determine a likelihood of collision between the first agent and each other agent, based on the initial estimated trajectory for the first agent and the initial estimated trajectory for each other agent;
• and
• generate a second estimated trajectory for the first agent based on the observed states of each of the plurality of agents, the map data, and at least one result of the first collision assessment.

A non-transitory computer readable medium embodying computer program instructions, the computer program instructions configured so as, when executed on one or more hardware processors, to implement operations comprising:

• receiving a set of training data comprising a set of training inputs, each input comprising an observed state of each of a plurality of agents, and map data of a mapped area, and a corresponding ground truth outputs, each ground truth output comprising an actual trajectory from the observed state of a first agent of the plurality of agents;
• processing the observed state of each agent and the map data in a first neural network to generate an initial estimated trajectory for each of a plurality of agents;
• performing a collision assessment to determine a likelihood of collision between the first agent and each other agent, based on the initial estimated trajectory for the first agent and the initial estimated trajectory for each other agent;
• processing the observed states of each of the plurality of agents, the map data, and the collision assessment results in a second neural network to generate a second estimated trajectory for the first agent;
• and
• computing an update to parameters of each of the first and second networks of the based on a loss function that penalises deviations between the initial estimated trajectory for the first agent and the corresponding ground truth output for the first agent, and deviations between the second estimated trajectory for the first agent and the corresponding ground truth output for the first agent.