Prediction
prediction inputs are: obstacles from perception module nad localization from localization module. outputs are obstacles will predicted trajectories.
there are three classes in prediction modules:
* container
store input dat from subscribed channelds, e.g. perception obstacles, vehicle localization
* evalutor
predicts paths and speed separately for any given obstacles
* predictor
generate predicted trajectories for obstacles. e.g. lane sequence(obstacle moves following the lanes), free movement, regional moves(move in a possbile region)
|
|
Prediction interface has defined: RunOnce() and Pulish(). the Prediction class has defined listener’s callback: OnLocalization
, OnPlanning
, and Start()
, Stop()
and implemented interface functions. For both EvaluatorManager and PredictorManager, there are Init()
and Run()
functions, and there are bunch of apis from container class.
Evaluator
|
|
and there are a few different evaluators: (multilayer perception approach) MLP and RNN(deep neural network), both will discuss in details in future.
Predictor
predictor will generate trajectories, a few apis:
|
|
the predictor has a few types:: regional based, free move, lane sequence e.t.c, which defined as subclasses, will discuss more in future.
Container
container manager class has a few subclass as adc_trajctory container, obstacles contianer and pose constainer, which should be discussed in details later.
Adapter
|
|