Tesla sales in China shock analysts!

[Leongsam]

 

[syed putra]

It's simple really.
Cities like Shanghai imposed RMB100k penalty to those buying ICE cars and nothing to those buying EV's. By imposing just this caused ICE car sales to drop and become uncompetitive.

 

[Leongsam]

 

[winners]

In reality and eventually worldwide, TESLA will not be able to beat BYD. People will want an economical EV and BYD's reliability is also improving year after year.

 

[Leongsam]

In reality and eventually worldwide, TESLA will not be able to beat BYD. People will want an economical EV and BYD's reliability is also improving year after year.

Tesla will win because of the superiority of its autonomous driving.

 

[winners]

Tesla will win because of the superiority of its autonomous driving.

Just tell me when will their RoboTaxis be approved and allowed on the roads?

When others are using LIDAR and RADAR for detection, TESLA chose to use only Images. Albeit cheap but is it reliable and dependable, especially during bad weather conditions?

 

[Leongsam]

Just tell me when will their RoboTaxis be approved and allowed on the roads?

My estimate is 2026 in some US cities. No idea when it will be allowed to go international.

 

[Betsycrant]

I think Tesla's camera system works well in most conditions, but bad weather is a real challenge they need to solve.

 

[eatshitndie]

My estimate is 2026 in some US cities. No idea when it will be allowed to go international.

robotaxis are launched in the entire state of arizona.

 

[Eisenhut]

Tesla will win because of the superiority of its autonomous driving.

Relying on pure camera vision is not superior

 

[syed putra]

Relying on pure camera vision is not superior

I was looking at the logic if there are too many inputs such as radar, vision, sensors etc will cause the car to be immobile or drive too slow to compute all the data.

 

[Eisenhut]

I was looking at the logic if there are too many inputs such as radar, vision, sensors etc will cause the car to be immobile or drive too slow to compute all the data.

There is no such things as "too much info"

All data bandwidth should be according to design specifications

 

[404]

Tesla Sales in China Are Popping, But That Means Nothing If You Can’t Stomach TSLA Stock’s Absurd P/E Ratio at 149

 

[Leongsam]

Relying on pure camera vision is not superior

In order to mimic human driving on a road system designed from the ground up based upon the vision provided by two human eyes a combination of AI and vision alone is the only way forward.

Tesla tried combining radar and vision and the results were inferior as conflicting inputs froze the decision making process.

 

[Leongsam]

There is no such things as "too much info"

All data bandwidth should be according to design specifications

It is not a case of "too much info". The problem was conflicting info.

From Grok:

Tesla's Full Self-Driving (FSD) system was intentionally designed around a camera-vision-only approach(often called "Tesla Vision") for philosophical, technical, scalability, and economic reasons. This is a deliberate bet rooted in first-principles thinking from Elon Musk and the Tesla AI team, rather than a compromise.

Core Philosophy: Mimic Human Vision​

Tesla argues that humans drive safely using primarily two eyes (biological cameras) and a brain, without radar or lidar. If biological vision suffices for complex, real-world driving, then advanced computer vision powered by neural networks should eventually do the same — and better.

Elon Musk has repeatedly called lidar a "crutch" and a "fool's errand," arguing that relying on it creates a false sense of progress and diverts effort from solving the fundamental problem of visual intelligence. Once you truly master vision, extra sensors become unnecessary.

This aligns with Tesla's AI-first approach: train massive neural nets on vast real-world video data to build generalized intelligence, similar to how large language models work.

Key Reasons for Rejecting Multi-Sensor Fusion (Radar + Lidar)​

1. Avoiding Sensor Contention and Ambiguity
   When sensors disagree (e.g., camera sees one thing, radar/lidar another), the system must resolve conflicts in real time. This can lead to hesitation, errors, or overly conservative behavior. Tesla views multi-sensor fusion as introducing more problems than it solves, especially as vision improves. A single, coherent "source of truth" (vision) simplifies decision-making.
2. Scalability and Data Advantages
   • Cameras are cheap, ubiquitous, and high-resolution: They provide rich, color, contextual data at high frame rates (like human eyes). Neural nets excel at extracting depth, velocity, semantics, and predictions from this.
   • Fleet-scale data: Tesla has millions of vehicles collecting billions of miles of video data. This "data engine" lets them rapidly iterate and improve the AI. Radar/lidar data is harder to scale universally and less rich for training general intelligence.
   • Future-proofing: Software improvements (neural net updates) can be pushed over-the-air to the entire fleet without hardware changes. Adding lidar/radar would require retrofitting or new hardware, limiting scalability.
3. Cost and Manufacturing
   Lidar sensors are expensive (especially high-quality ones), increasing vehicle cost and complexity. Radar adds cost too. Tesla's approach keeps vehicles more affordable and simpler to produce at volume, which supports their goal of widespread robotaxi deployment. Cameras + powerful onboard compute (like the HW3/HW4/AI5 chips) is seen as the economical path to autonomy.
4. Performance Trajectory
   Tesla phased out radar around 2021 (starting with Model 3/Y in North America), moving fully to vision because they believed neural nets had surpassed radar's usefulness in many scenarios. Vision handles dynamic environments, object recognition, and edge cases better as models improve, while radar can introduce noise or limitations (e.g., in certain weather or for semantic understanding).

Evolution of Tesla's Sensor Suite​

• Early Autopilot: Cameras + radar + ultrasonics.
• 2021 Pivot (Tesla Vision): Removed radar from new vehicles; focused on pure vision. Ultrasonics were also later minimized or removed in some models.
• Current FSD: Primarily 8+ cameras, with heavy reliance on end-to-end neural networks (planning and control directly from vision inputs).

Tesla has acknowledged short-term challenges during transitions (e.g., temporary limitations when dropping radar), but they maintain vision-only will ultimately be safer and more capable.

Counterpoints and Context​

Critics (and competitors like Waymo, Cruise) argue that multi-sensor redundancy improves safety in edge cases (fog, heavy rain, low light, glare), where cameras can struggle. Lidar excels at precise 3D mapping and distance. Tesla counters that these are "crutches" that delay true AI development, and that a well-trained vision system + prediction can handle them. Real-world results are mixed and evolving — FSD has improved significantly but still requires supervision and faces regulatory scrutiny.

In summary, Tesla's camera-only design is a high-conviction strategy betting on AI progress over sensor hardware redundancy. It prioritizes simplicity, cost, data scale, and human-like intelligence over a "more sensors = safer" approach. Whether it proves superior long-term remains one of the biggest debates in autonomous driving.

 

[Leongsam]

Time will tell who is correct.

 

[Eisenhut]

It is not a case of "too much info". The problem was conflicting info.

From Grok:

Tesla's Full Self-Driving (FSD) system was intentionally designed around a camera-vision-only approach(often called "Tesla Vision") for philosophical, technical, scalability, and economic reasons. This is a deliberate bet rooted in first-principles thinking from Elon Musk and the Tesla AI team, rather than a compromise.

Core Philosophy: Mimic Human Vision​

Tesla argues that humans drive safely using primarily two eyes (biological cameras) and a brain, without radar or lidar. If biological vision suffices for complex, real-world driving, then advanced computer vision powered by neural networks should eventually do the same — and better.

Elon Musk has repeatedly called lidar a "crutch" and a "fool's errand," arguing that relying on it creates a false sense of progress and diverts effort from solving the fundamental problem of visual intelligence. Once you truly master vision, extra sensors become unnecessary.

This aligns with Tesla's AI-first approach: train massive neural nets on vast real-world video data to build generalized intelligence, similar to how large language models work.

Key Reasons for Rejecting Multi-Sensor Fusion (Radar + Lidar)​

1. Avoiding Sensor Contention and Ambiguity
   When sensors disagree (e.g., camera sees one thing, radar/lidar another), the system must resolve conflicts in real time. This can lead to hesitation, errors, or overly conservative behavior. Tesla views multi-sensor fusion as introducing more problems than it solves, especially as vision improves. A single, coherent "source of truth" (vision) simplifies decision-making.
2. Scalability and Data Advantages
   • Cameras are cheap, ubiquitous, and high-resolution: They provide rich, color, contextual data at high frame rates (like human eyes). Neural nets excel at extracting depth, velocity, semantics, and predictions from this.
   • Fleet-scale data: Tesla has millions of vehicles collecting billions of miles of video data. This "data engine" lets them rapidly iterate and improve the AI. Radar/lidar data is harder to scale universally and less rich for training general intelligence.
   • Future-proofing: Software improvements (neural net updates) can be pushed over-the-air to the entire fleet without hardware changes. Adding lidar/radar would require retrofitting or new hardware, limiting scalability.
3. Cost and Manufacturing
   Lidar sensors are expensive (especially high-quality ones), increasing vehicle cost and complexity. Radar adds cost too. Tesla's approach keeps vehicles more affordable and simpler to produce at volume, which supports their goal of widespread robotaxi deployment. Cameras + powerful onboard compute (like the HW3/HW4/AI5 chips) is seen as the economical path to autonomy.
4. Performance Trajectory
   Tesla phased out radar around 2021 (starting with Model 3/Y in North America), moving fully to vision because they believed neural nets had surpassed radar's usefulness in many scenarios. Vision handles dynamic environments, object recognition, and edge cases better as models improve, while radar can introduce noise or limitations (e.g., in certain weather or for semantic understanding).

Evolution of Tesla's Sensor Suite​

• Early Autopilot: Cameras + radar + ultrasonics.
• 2021 Pivot (Tesla Vision): Removed radar from new vehicles; focused on pure vision. Ultrasonics were also later minimized or removed in some models.
• Current FSD: Primarily 8+ cameras, with heavy reliance on end-to-end neural networks (planning and control directly from vision inputs).

Tesla has acknowledged short-term challenges during transitions (e.g., temporary limitations when dropping radar), but they maintain vision-only will ultimately be safer and more capable.

Counterpoints and Context​

Critics (and competitors like Waymo, Cruise) argue that multi-sensor redundancy improves safety in edge cases (fog, heavy rain, low light, glare), where cameras can struggle. Lidar excels at precise 3D mapping and distance. Tesla counters that these are "crutches" that delay true AI development, and that a well-trained vision system + prediction can handle them. Real-world results are mixed and evolving — FSD has improved significantly but still requires supervision and faces regulatory scrutiny.

In summary, Tesla's camera-only design is a high-conviction strategy betting on AI progress over sensor hardware redundancy. It prioritizes simplicity, cost, data scale, and human-like intelligence over a "more sensors = safer" approach. Whether it proves superior long-term remains one of the biggest debates in autonomous driving.

Micmic human vision is OK. But using only Human Micmic vision is wrong.

Because even human beings dont depend on vision only.....we have sound, feel temp, wind...etc