Tesla’s Full Self-Driving (Supervised) software has made remarkable strides with end-to-end neural networks, but the engineering roadmap ahead is even more exciting. Instead of simply following road lines and navigating traffic lights, future updates could overhaul how drivers interact with their vehicles by transforming FSD into a system that communicates naturally, respects personal driving choices, and (finally) learns to avoid potholes.
Between major architectural upgrades and new hardware, several new features are lined up for upcoming releases. Here’s an in-depth look at everything we currently know is coming to FSD.
Grok-Powered FSD Voice Navigation
One of the biggest pain points with voice controls in modern cars is how rigid the prompts feel. Tesla has been gradually changing that with Grok, letting drivers converse naturally with the AI assistant to set location-based reminders, configure their driving route, and even control cabin features and settings toggles like climate and steering weight. Tesla plans to take things even further by embedding SpaceXAI’s chatbot directly into the driving stack, with FSD getting support for Grok voice controls to handle multi-step, contextual commands.
Instead of being able to follow more chauffeur-like phrases such as “take the next right,” we expect Grok and FSD to work in tandem to understand high-level intent. You could say, “Navigate to the grocery store through the downtown strip, avoid Main Street construction, and park in the closest stall,” and the car will interpret the full request. This would include understanding specialized parking choices, such as pulling into accessible bays when authorized or finding covered parking on hot days.
Pothole Avoidance
Craters and broken road surfaces remain a constant annoyance for drivers using FSD, often forcing manual takeovers to prevent an uncomfortable impact and wear on tires and wheels. Fortunately, Elon Musk recently confirmed pothole avoidance is coming to FSD to help vehicles steer clear of damaged asphalt.
Tesla previously used map and fleet data to soften the active air suspension on Model S and Model X when approaching rough road segments. The upcoming visual model will go much further by identifying dips and broken surfaces in the camera feed in real time, smoothly nudging the vehicle around them while staying inside the lane.
Mirroring Your Personal Parking Habits
Autopark is effective at sliding between painted lines, but it lacks the nuance of an experienced driver. An upcoming update will allow FSD to copy your parking habits, learning individual preferences over time.
Whether you prefer backing into home driveways instead of going nose-first, tucking extra close to garage walls to leave walking space, or positioning wheels away from curbs, FSD will analyze manual park jobs and try to replicate those placements when parking autonomously.
Remembering Driver Interventions
When navigation insists on taking a route you dislike, constantly fighting the wheel gets tiresome. Tesla is solving this by ensuring FSD will remember driver interventions and route preferences.
This ties directly into the recently deployed Preferred Routes feature, which began allowing FSD to learn the driver’s favorite commuter shortcuts, and could even be part of it. Moving forward, repeated steering nudges or manual lane choices will serve as positive feedback, training your vehicle to stick with the specific roads and turns you routinely choose.
FSD v15: A Massive Parameter Leap
While v14 builds continue rolling out, Tesla’s AI team is hard at work preparing FSD v15. The next generational jump represents a tenfold parameter increase, expanding from around one billion parameters to a massive ten-billion-parameter neural network.
The expanded architecture focuses on earlier hazard prediction, faster reaction speeds, and better collision avoidance, among other improvements. FSD v15 is expected to debut sometime later this year or early next year, and it also stands to improve Automatic Collision Evasion, the latest FSD feature Tesla released.
Hardware Roadmap: AI4+ Upgrades and AI5
Software can’t advance indefinitely without increments in raw compute power. For older Hardware 3 cars, Tesla announced an upgraded AI4+ (or AI4.5) computer with expanded memory.
With FSD v14 Lite likely marking the final major branch for HW3 vehicles, this refreshed computer is widely expected to be the promised retrofit path required to give legacy cars enough processing muscle for unsupervised autonomy. Meanwhile, Tesla’s next-generation 2-nanometer AI5 chip has already started trial production in Texas, on track for high-volume manufacturing in 2027. However, AI5 will initially power Optimus humanoid robots and Tesla’s data centers before eventually migrating to consumer cars.
Together, these upcoming software and hardware updates show that Tesla is shifting its focus from basic autonomous steering to building an intuitive, personalized co-pilot.
