Closed davidhopper2003 closed 5 years ago
We are using data to trigger each module to minimize the end-to-end time delay. You can add a dummy prediction to provide some prediction topic so that the planning can be triggered.
@lianglia-apollo I've got it. Thank you very much.
Closing this issue as it appears to be resolved. @davidhopper2003 feel free to reopen it if you have additional questions. Thanks!
@lianglia-apollo , previously with the timer-trigger scheme, if the perception/prediction module gets stuck, planner module can still compute the trajectory using the latest localization data. Now we lose this property with the new data-trigger design. How do you consider this problem when you make this design change?
Under the Cyber RT framework, the update of the planning module is changed from the timer triggering mechanism to the message updating mechanism. In other words, a specific planner such as NaviPlanner of the planning module only works (via
planning_base_->RunOnce (local_view_, &adc_trajectory_pb);
) in the case where all three messagesprediction::PredictionObstacles
,canbus::Chassis
andlocalization::LocalizationEstimate
have arrived. It causes a problem for debugging the planning algorithm. Usually, a recorded data bag file only contains two types of messages:canbus::Chassis
andlocalization::LocalizationEstimate
; theprediction::PredictionObstacles
messages doesn't exist in general. If we start thePerception
andPrediction
modules to generate such a message, the following core dump message will appear:Why the planning module can't be triggered based on a timer to facilitate the debugging of a specific planning algorithm?
System information