Asleep At The Wheel: Tesla's Autopilot
In recent years, autonomous driving technology has been controversial due to safety concerns. Contrary to popular belief, Tesla’s vehicles are not autonomous.
However, there are no self-driven vehicles in the market. Earlier this year, Tesla CEO, Elon Musk, talked about how there are many challenges to achieve perfect self-driving technology. “So I guess like, I don't know, it’s 98% good right now, but we need it to be like 99.999%”. The between 98% and 99.999% is everything. For example, if you drive a car for 100 hours then there are two hours the Tesla cannot ensure your safety, and you may get into an accident. If people wholeheartedly believe in autopilot technology, it could cause serious safety problems.
Deep Learning for Improving
Tesla’s Autopilot is based on deep learning technology. The firm created one of the most significant, complex neural networks in the world. They believe they cannot solve most of the problems with self-driving technology without using neural networks, but they need overwhelming data to optimize the models.
In the meantime, the sales of Tesla have skyrocketed in the past two years, meaning they can get mass data for training and testing the network models. The drive system can improve every day by updating the driving algorithm. As a result, Tesla may achieve perfect autonomous driving in the near future.
How to Process the data? Software 2.0
Andrej Karpathy, who is the director of artificial intelligence and Autopilot Vision at Tesla, stated that Tesla wants to use Software 2.0 to help them to process mass data. Software 2.0 means Machine Learning can take the place of human to design and create applications. Take neural networks as an example. Software 2.0 can help people to design more complex and abstract networks. It also can explain some features that human cannot explain, like the weights of neural networks.
Software 2.0 can also solve the ‘label’ problems. After Tesla collects the data, most of them are unlabeled. Tesla can use Software 2.0 to cluster the data with the data engine. To state data engine more clearly, it is a process about labeling the data to the database, training them for the models, deploying, noticing a problem, boosting, and doing it again and again. Tesla can improve models efficiency and reduce labor costs.
However, it is just the beginning of Software 2.0, and there are no companies tried it before. Tesla has already created a significant amount of tooling that assists humans in writing 1.0 code. Tesla’s deep learning technology will gradually become software 2.0, but it will take some time.
Written by Zhijun Qiu & Edited by William Turchetta & Alexander Fleiss