Motion policy planner for navigation

A system comprising one or more processors to:

• generate perception data corresponding to a representation of at least a portion of an environment that includes at least an ego agent and one or more peripheral agents;
• and
• using one or more machine learning models, generate a motion policy for navigating the ego agent through the environment based at least on the perception data, wherein to generate the motion policy, the one or more machine learning models are trained to predict a policy tree structure corresponding to a predicted output of a gradient-based optimization configured to optimize a trajectory of the ego agent based at least on a concurrent optimization of trajectories of the peripheral agents, the optimization being further based at least on the perception data.

A method comprising:

• generating, using one or more machine learning models, a motion policy for navigating an ego machine through an environment based at least on the one or more machine learning models predicting an output representing a policy tree computed using a gradient-based optimization algorithm that optimizes a trajectory of the ego machine based at least on a concurrent optimization of trajectories of the peripheral agents in response to an input comprising at least one input of environmental perception data associated with the ego machine.