Mobileye CEO Amnon Shashua has sparked debate after arguing against Tesla’s end-to-end deep learning approach to full self-driving capability.
In a blog post on X, Shashua contends a purely data-driven end-to-end FSD V12 lacks necessary transparency, controllability, and proven performance for truly driverless autonomy.
Elon Musk’s FSD Beta v12 Demo Showcases Major AI Breakthrough for Self-Driving
The post comes after Tesla indicated its upcoming FSD beta v12 version will rely more heavily on neural networks versus a highly engineered modular system like Mobileye’s SuperVision. VW has collaborates with Intel-owned Mobileye on lidar, radar, cameras and software for autonomous functionality up to SAE Level 4.
But Shashua believes Mobileye’s hybrid architecture combining neural networks into a rules-based system is superior for safety-critical applications. He argues fully neural approaches can make unpredictable errors unacceptable for passenger vehicles.
While acknowledging the power of deep learning, Shashua states it must be carefully incorporated within a transparent, verifiable framework. Mobileye’s approach applies targeted AI while retaining regulatory visibility into decision-making.
Shashua also took aim at the high cost of Tesla’s FSD, which he feels is unjustified by the capability delivered so far. In contrast, he claims Mobileye’s SuperVision hardware could reach full autonomy for around $1,000 in costs.
The criticism highlights growing skepticism from sector veterans around Musk’s autonomous promises. But Tesla remains confident its data-driven path can unlock full self-driving safer and sooner than the incremental approaches of legacy players.
The two competing schools of thought both face immense challenges on the road to scalable driverless transport. But for now, should Mobileye successfully solve Full Self-Driving, it would lend more credibility to the CEO’s statements. Despite not having attempted it himself, Amnon has criticized the Tesla end-to-end FSD v12.