Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revision Previous revision
Next revision 		Previous revision
Member:sungbeanJo_paper [2021/03/05 12:36]
sungbean 		Member:sungbeanJo_paper [2021/04/21 22:08] (current)
sungbean
Line 1: Line 1:
-Behavioral cloning solves a regression problem in which +get_config_param active timestamp_mode  
-the policy parameterization is obtained by maximizing the + TIME_FROM_INTERNAL_OSC 
-likelihood of the actions taken in the training data. BC works +get_config_param active multipurpose_io_mode 
-well for states adequately covered by the training data. It + OUTPUT_OFF ​ 
-is forced to generalize when predicting actions for states +get_config_param active sync_pulse_in_polarity 
-with little or no data coverage, which can lead to poor + ACTIVE_LOW 
-behavior. Unfortunately,​ even if simulations are initialized +get_config_param active nmea_in_polarity 
-in common states, the stochastic nature of the policies allow + ACTIVE_HIGH 
-small errors in action predictions to compound over time, +get_config_param active nmea_baud_rate 
-eventually leading to states that human drivers infrequently + BAUD_9600 
-visit and are not adequately covered by the training data. +
-Poorer predictions can cause a feedback cycle known as +
-cascading errors [20]. +
-In a highway driving context, cascading errors can lead +
-to off-road driving and collisions. Datasets rarely contain +
-information about how human drivers behave in these situations,​ +
-which can lead BC policies to act erratically when +
-they encounter such states. +
-Behavioral cloning has been successfully used to produce +
-driving policies for simple behaviors such as car-following +
-on freeways, in which the state and action space can be +
-adequately covered by the training set. When applied to +
-learning general driving models with nuanced behavior and +
-the potential to drive out of lane, BC only produces accurate +
-predictions up to a few seconds [5, 6].+

Trace: • SungilKim

Navigation