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 [2021/03/04 18:03]
sungbean
Member:sungbeanJo_paper [2021/04/21 22:08] (current)
sungbean
Line 1: Line 1:
-We use L1 as loss function ℓ instead +get_config_param active timestamp_mode  
-of the mean squared error (MSE), as it is more correlated + TIME_FROM_INTERNAL_OSC 
-to driving performance [11]. As our NoCrash benchmark +get_config_param active multipurpose_io_mode 
-consists of complex realistic driving conditions in the presence + OUTPUT_OFF ​ 
-of dynamic agents, we collect demonstrations from an +get_config_param active sync_pulse_in_polarity 
-expert game AI using privileged information to drive correctly + ACTIVE_LOW 
-(i.e. always respecting rules of the road and not crashing +get_config_param active nmea_in_polarity 
-into any obstacle). Robustness to heavy noise in the + ACTIVE_HIGH 
-demonstrations is beyond the scope of our work, as we aim +get_config_param active nmea_baud_rate 
-to explore limitations of behavior cloning methods in spite + BAUD_9600 
-of good demonstrations. Finally, we pre-trained our perception +
-backbone on ImageNet to reduce initialization variance +
-and benefit from generic transfer learning, a standard practice +
-in deep learning seldom explored for behavior cloning.+
Navigation