Why Autopilot 2.0 is a Big Deal
To fully appreciate Autopilot 2.0, one must understand the limitations of Autopilot 1.0. Autopilot 1.0 was built upon Mobileye’s object recognition system, which was based on a low-power GPU and rigid rules for object recognition. In some cases, Mobileye’s system would/could not recognize certain objects and this posed a serious limitation (and danger) to Tesla’s Autopilot system. Tesla built a complex and very capable system (that included HD mapping, radar analysis, and driving decision making) on top of Mobileye’s object recognition system. But the flaws of Mobileye’s system could only be mitigated by Tesla and not overcome. In order to truly improve Autopilot in a substantial manner, Tesla needed to develop their own object recognition system and own the entire software stack of autonomous driving. This is what they did for Autopilot 2.0.
Put simply, Autopilot 2.0 takes the input from cameras, radar, sensors and processes them on a high-powered GPU and computer. The key is Tesla is not using rigid rules (that have limitations) but rather they’re using a "neural net" (a type of AI system), which continuously gets better and will easily be 10x (if not 100x) more accurate than their previous system. (note: for more info on neural networks, see link below in Deep Learning Simplified)
As the neural net gets trained, it becomes increasingly accurate. The process should be fairly quick, and Autopilot 2.0 should be more accurate than Autopilot 1.0 within a 3-4 months (my guess). Within 6-12 months, Autopilot 2.0 will likely be 2-3x more capable and accurate than Autopilot 1.0. This is a big deal because it’s safety we’re dealing with, and Autopilot 2.0 will be a substantially safer experience.
The reason why Elon thinks Autopilot 2.0 will reach full self-driving capabilities within a couple years is because he’s confident in the power of the neural net (and the power of AI) to make itself better. In Elon’s words, this is a “solved problem.” Regulatory approval might take longer, depending on the country/jurisdiction, but the technology is at our doorstep.