Will Tesla Vision Be Enough to Enable a Full Self-Driving Future?
Tesla began transitioning its fleet over to Tesla Vision towards the middle of last year in a bold move signifying that they will be able to enable autonomy solely with a sophisticated camera system: will this ambitious bet achieve level 5 autonomy?
Previously, Tesla had built every Tesla since November 2016 with dedicated Full Self-Driving hardware which consisted of eight cameras, forward-facing radar, and twelve sonars. Tesla began its transition to Tesla Vision by removing the forward-facing radar from Model 3 and Model Y, while the sonars and cameras remain.
It’s a move that was detested by autonomous experts, despite being suggested directly by Elon Musk. His justification was that as humans are able to drive with two eyes, cars should be able to drive without radar. This caused criticism by numerous agencies, which backtracked once they were able to test a new version of Autopilot and its various safety features utilizing the Tesla Vision system. The IIHS awarded both versions of Tesla’s Autopilot collision avoidance a “Superior” rating.
As we’ve learned yesterday, Tesla Vision isn’t perfect or without flaws for collision avoidance as compared to the previous system. At the same time, it’s relatively new and Tesla has been selling more vehicles than ever which increases the sample size.
Above all Tesla Vision is forced to rely on deep neural net training, which is an incredibly complex system that’s essentially an advanced cloud-connected supercomputer.
Tesla hosted an AI Day event outlining what they’ve accomplished and what further work there is for Full Self-Driving to be successfully powered by a neural net, including an in-house Dojo learning computer. There is a large, capable team working on sophisticated algorithms that can identify and map objects as well as predictively assign them their own moving identifier in space.
It makes the use of radar almost seem redundant. We have to raise the question of whether removing radar provides a superior experience, and we’d conclude that to be a resounding ‘no’ at the surface level. At the same time, it’s making Tesla work harder at developing the most advanced neural net of any automaker or company in the world.
A lack of radar can still provide a comparable experience, with a proper neural net. An issue that arises is that there’s less of a failsafe where the system can cross-check against the radar to ensure that an object is or isn’t within the vehicle’s direct path.
If it isn’t truly necessary, then that amounts to significant cost savings over time. Tesla is planning to begin building millions upon millions of vehicles per year within the next decade, and every vehicle built will have the Full Self-Driving hardware suite. Thus, even a ~ $100 radar component provides an area where Tesla can cut cost for both the consumer and their bottom line.
There is still evidence to the contrary which signifies that Tesla may not be ready to make a full commitment: Model S and Model X still ship with radar, despite being refreshed last year and shipping in the summer coinciding with Tesla’s switch over to Tesla Vision on Model 3 and Model Y.
Tesla’s own explanation is that as Model 3 and Model Y are higher volume vehicles, they provide the automaker with a larger volume of real-world data that they can then analyze and label for use within the neural net. Model S and Model X should follow eventually.
The proof that radar may not be necessary: the Full Self-Driving Beta operates without the use of radar regardless of whether a vehicle is equipped with it.
As the Beta has been getting substantially better over time, Tesla may not be wrong that Tesla Vision is all that’s required. While it can still make the wrong moves at the wrong time, testers have been finding that it has required less interventions and in some cases can even travel from point A to point B uninterrupted.
On the other hand, what we’ve found is that radar ultimately doesn’t hurt as it provides depth data that even the best processing in the world can’t reliably replicate all of the time. It can get shockingly close, but sonar and cameras alone can’t track objects quite as well. The only true downside is cost as the average vehicles uses thousands of parts regardless, and from that perspective it is possible to further reduce cost with volume and engineering: there are cheaper alternatives available to automakers.
Yet, Tesla doesn’t find radar to be essential to Full Self-Driving.
When we consider the levels of autonomy, Tesla may have a point: radar is more essential for basic driver assistance features such as active cruise control which are only looking at the road ahead up to level 3.
Once we get up to level 4, which is Full Self-Driving in most conditions and arguably where the Full Self-Driving Beta is approaching, then a much more complex system is needed which Tesla is well into the process of building. These complicated systems replace more basic hardware by considering and predicting all aspects of a driver’s surroundings, rather than simply distance or lane markings.
Level 5 full autonomy without interventions should hypothetically be possible with Tesla Vision, as the team is working diligently to make that a reality utilizing neural net technology that’s significantly more advanced than the previous-generation camera/radar system designed by Mobileye which was utilized for Autopilot 1.0.
As Tesla is looking to the future, they’re developing the most forward-thinking system on the market without the use of radar and we don’t find it to be a necessary component as what they’re creating with neural net training is significantly more advanced — which is exactly what level 4 and subsequently level 5 autonomy require.
Tesla Vision should be what’s necessary for a Full Self-Driving future. While it may not be the final generation, as hardware will evolve over time, it should be the first time we see at the least level 4 autonomy achieved on a consumer product by an automaker.
__________________
Interested in Listing or Upgrading YOUR Tesla? The BEST place to get TOP dollar for your used Tesla. MADE IN NYC
Tesla currently has limited new inventory available for immediate purchase, and used vehicles across the automotive industry are selling for an average of $5,000 over market value.
Browse the listings now: https://onlyusedtesla.com/listings/
Or feel free to contact us directly at contact@onlyusedtesla.com today!
