So we've been testing...a lot.
TL;DR
We've made sure the system is easily installable. We've tested the machine learning model to make sure it's accurate and have made some tweaks. We've also taken this out on the road and seen some very good results.
What Happened
We recorded how to install the system - this involved disassembling and reassembling. This confirmed that the system was easy to install. We clocked in at under 15 minutes and almost zero physical modifications to the car.
We also did a lot of testing. We took the system out for a few drives to collect our test set data then annotated all of it. Our annotations included drawing boxes around the hazards and labelling them. We could now test our model and gain an actual numeric metric showing us how good the model is. We're currently sitting at around 99% accuracy but this is somewhat optimistic because all the data we have is in the day time in perfect clear weather conditions. We will be collecting more data as the weeks go by, including nighttime data, and if possible, snow and rain.
What's Next
More data collection, more testing. We'll also be recording our marketing and demo videos.
Comments