Tesla Electric Vehicle Tesla Model S claimed its first victim, ironically a Tesla diehard fan.On may 7 2016, the Tesla he was driving on Autopilot,ended up under the trailer crossing the road perpendicular to Tesla.
What went wrong with the auto pilot. Tesla puts the blame partially to itself and partially to the driver.Autopilot failed to notice the trailer against the backdrop,which is a correct assumption (excuse). Did the driver put faith on the system too much , some reports say that he was playing harry potter movie.Driving on a Six lane straight road, has become nontrivial for a Tesla.May his soul rest in peace, he died doing what he loved ,Tesla and Harry Potter.
Computer vision relies on Bayer Array Filter Digital Sensor. Digital Sensor are never going to be able to match our eyes , What makes them superior is because of the processing that is done on them. And the Algorithms go wrong all the time.
Machine Learning to the rescue.
Machine learning by definition is machine learning something. The machine in question is Tesla Autopilot from a Company called Magic eye. The machine is learning somuch for so long that ,if it was a person , you will never be able to meet him . The inherent issue here is that it is learning on and on.
Learn from other's mistake.
There is so much for it to learn. Search on YouTube for DashCam Crash Compilation. What do you see, you see people doing mostly the same mistakes but also something completely different that it makes you laugh. I don't know if Tesla feeds these videos onto its system,but it should. The data collected from the Tesla drivers is carefully driven journeys comprising Movement of slight deviances,missing markers,curves etc. What's missing are Hardcore Data. The serious data of crashes per 96 million miles.
Data Data More Data
Data needs to be from a wide range of vehicles, all vehicles if possible. Google testing self driving cars ,closed to public before becoming mainstream has the same problem as Tesla. It lacks the data, Varied , Noisy Data to be precise. Make Strategic partnership with other manufacturers. Ultimately Share the data. Create devices to be retrofitted, like DashCam devices,Accelerometers,etc. to manual cars (Automatic is the new Manual Now!)
The regulatory body may ask Tesla to disable Autopilot until ready or just your jealous neighbour might lobby for that. But it cannot be ready without ground check. The year old discussion between educators,Theory vs Practical,comes into play. This quantically small accident may have an impact but like the horse cart vs cars, it will be short and it will be decisive on the superior side. Self Driven is the future, whether or not Electric,that will be answered politically.