Tesla limits advanced AI driving features to newer vehicles

The race to perfect autonomous driving capabilities continues as Tesla rolls out significant software updates, though access to the latest features remains hardware-dependent.

Latest deployment details: Tesla has released version 12.5.6.3 of its Full Self-Driving (FSD) software, introducing end-to-end neural networks for highway driving specifically for vehicles equipped with HW4 (AI4) hardware.

• The update extends neural network control across highways, city streets, and parking lots for compatible vehicles
• End-to-end neural networks allow AI to control the vehicle directly from visual input rather than relying on explicitly coded instructions
• The release was initially planned for October but faced delays before its wider rollout in November

Technical specifications: End-to-end neural networks represent a significant advancement in how Tesla’s autonomous driving system processes and responds to road conditions.

• The system now uses neural networks to handle all driving scenarios, moving away from separate software stacks for different driving environments
• Tesla’s head of self-driving and AI, Ashok Elluswamy, confirms the update includes customizable driving styles for speed and lane change preferences
• The technology promises smoother, more natural, and potentially safer highway driving compared to previous versions

Hardware limitations: A clear divide is emerging between Tesla’s newer and older vehicles in terms of autonomous driving capabilities.

• HW3 (AI3) vehicles are restricted to “improved v12.5.x models for city driving” according to Tesla’s roadmap
• The company has acknowledged reaching computational limits with the HW3 hardware
• No retrofit options have been announced for owners of vehicles with older hardware
• Millions of HW3 vehicle owners are currently excluded from accessing the latest end-to-end neural network features

Real-world implications: Early user experiences with recent FSD updates highlight ongoing challenges with the technology’s reliability and consistency.

• Some users report regression in performance with version 12.5.4.2 compared to previous releases
• Highway exit handling remains problematic, with instances of unsafe maneuvering reported
• The lack of hardware upgrade paths for older vehicles raises concerns about long-term support and improvement potential

Future outlook: Tesla’s diverging development paths for different hardware versions suggests a potentially permanent split in feature availability, raising questions about the company’s commitment to its earlier promises of universal self-driving capabilities through software updates alone. The absence of retrofit options may ultimately force owners of older vehicles to upgrade to newer models to access the latest autonomous driving features.