agent

An agent receives measurements of sensors from the physics engine and gives a control input by a combination of various algorithms.

agent module defines different types of computation paradigm, such as ModelBasedAgent and LearningBasedAgent. Each type defines a way to combine algorithms. For example, ModelBasedAgent always uses model, estimator, planner, and controller. Where RLBasedAgent only uses RL_controller module.

A user can build an agent by specifying which algorithms to use and hyperparameters of algorithms. The algorithms may have dependencies on other algorithms, all the required algorithms should also be used.

agent connects algorithms with each other. For example, the planned trajectory given by planner is composed of multiple waypoints. The waypoints are fed into the controlled one by one by agent.