Differences
This shows you the differences between two versions of the page.
Both sides previous revision Previous revision Next revision | Previous revision | ||
Member:sungbeanJo_paper [2021/03/04 16:33] sungbean |
Member:sungbeanJo_paper [2021/04/21 22:08] (current) sungbean |
||
---|---|---|---|
Line 1: | Line 1: | ||
- | Despite its positive performance, we identify limitations | + | get_config_param active timestamp_mode |
- | that prevent behavior cloning from successfully graduating | + | TIME_FROM_INTERNAL_OSC |
- | to real-world applications. First, although generalization | + | get_config_param active multipurpose_io_mode |
- | performance should scale with training data, generalizing | + | OUTPUT_OFF |
- | to complex conditions is still an open problem with a lot | + | get_config_param active sync_pulse_in_polarity |
- | of room for improvement. In particular, we show that no | + | ACTIVE_LOW |
- | approach reliably handles dense traffic scenes with many | + | get_config_param active nmea_in_polarity |
- | dynamic agents. Second, we report generalization issues | + | ACTIVE_HIGH |
- | due to dataset biases and the lack of a causal model. We | + | get_config_param active nmea_baud_rate |
- | indeed observe diminishing returns after a certain amount | + | BAUD_9600 |
- | of demonstrations, and even characterize a degradation of | + | |
- | performance on unseen environments. Third, we observe a | + | |
- | significant variability in generalization performance when | + | |
- | varying the initialization or the training sample order, similar | + | |
- | to on-policy RL issues [19]. We conduct experiments | + | |
- | estimating the impact of ImageNet pre-training and show | + | |
- | that it is not able to fully reduce the variance. This suggests | + | |
- | the order of training samples matters for off-policy Imitation | + | |
- | Learning, similar to the on-policy case [46]. | + |