There are a lot of misunderstandings, myths, even outright lies about Tesla’s FSD Beta that have been gaining traction and circulating around the internet. Since we view Tesla as a future disruptor of the automobile and transportation industry, it is important for investors to gain a clear understanding of what is happening and whether FSD Beta bodes well for Tesla’s future.
FSD Beta is one step towards fully autonomous driving
Tesla’s Full-Self Driving (FSD) Beta is its initial step towards fully autonomous driving. To state the obvious, FSD Beta is not itself autonomous driving, but a step towards autonomous driving.
The FSD Beta software release has one main purpose: to train Tesla’s artificial intelligence (AI) to drive safely in the real world by learning from the data gathered by good real world drivers. In other words, FSD Beta is primarily a data gathering tool for Tesla. Tesla learns from the good, the bad, and the ugly, in order to develop a driving computer that can safely drive in the real world.
FSD Beta was initially released in October 2020, then expanded to about 2,000 people in March 2021. This week, Tesla has moved towards a wider roll out by allowing drivers to request access to FSD Beta:
As seen above, the disclaimers are that: (1) Tesla will collect and review your driving data, (2) you remain responsible for staying alert with hands on the wheel, and (3) FSD Beta can be revoked at any time.
Opting in does not automatically grant access to FSD Beta; drivers are first assessed via Tesla’s Safety Score Beta as well as their existing driving data. If the Tesla team deems the driver safe enough, the driver will gain access to FSD Beta via a software update.
In a September 28 post, Elon Musk mentioned that the roll-out would begin Friday, October 8, 2021, with about 1,000 owners per day.
Note: Musk issued a correction in the comments
Safety Score Factors
This safety score is based on five factors (technical details are explained in Tesla’s explanation).
- Forward Collision Warnings: these are audible and visual alerts when a collision could occur in front of the car, usually because of unsafe following distance and/or erratic acceleration.
- Hard braking: defined as a decrease in speed greater than 6.7mph in one second, or backward acceleration in excess of 0.3g.
- Aggressive turning: defined as an increase in speed greater than 8.9mph in one second, or left/right acceleration in excess of 0.4g.
- Unsafe following distance (only applies when traveling 50mph or more): Based on the car’s own speed, the speed of the car in front, and the distance between the two cars -- the driving computer can calculate the number of seconds you would have to react and stop if the car in front came to a sudden stop. This measurement is called “headway.” Unsafe following is the proportion of time where your car’s headway is less than 1.0 seconds relative to the time that your car’s headway is less than 3.0 seconds. In other words, a 3 second cushion to stop the car suddenly is the ideal.
- Forced Autopilot Disengagement: when the car has determined that you have removed your hand from the steering wheel or that your eyes are inattentive (via the rearview mirror mounted camera), you will receive audio and visual warnings. After three warnings, Autopilot will disengage.
This data is compiled continuously and a score is given at the end of every trip. Daily scores (of up to 30 days) are combined into a mileage-weighted average to calculate your aggregated Safety Score, on a scale of 0 to 100.
This “test” has led to a wide variety of responses, mainly negative and critical:
- “Tesla Is Now Ranking Drivers With A 'Safety Score' And Of Course It's Already Going Weird. Unsurprisingly, judging driving behavior without context turns safety into a stupid game.” (Jalopnik)
- “Only a select number of FSD owners will get to use the beta – if their seemingly arbitrary safety scores measure up” (Electrek)
- “Tesla is putting untrained drivers on public roads as testers for their misleadingly-named, unproven system—a seeming recipe for disaster. Serious safety concerns should put this reckless plan in reverse. It’s Russian Roulette for unsuspecting drivers & the public.” (U.S. Senator Richard Blumenthal)
- “Tesla Starts Judging Owners It Charged $10,000 for Self-Driving” (Bloomberg)
Five Misunderstandings About Tesla's Safety Score and FSD Beta
Misunderstanding #1: Your Safety Score is the only factor in determining access to FSD Beta -- No!
Tesla is very clear that a combination of existing driving data and daily Safety Scores are used to determine eligibility for FSD Beta. Bad drivers who all of a sudden change their driving habits are easily identified based on their pre-Safety Score driving data. The Safety Score “test” is only one piece of the entire picture that the Tesla team looks at to assess the safety of a driver.
Drivers attempting to “hack” the system by driving uncharacteristically slow or overly cautious can be easily identified and rejected based on past driving data prior to opting into FSD Beta.
It seems fairly clear from Tesla’s explanation that the Safety Score and driving data are evaluated by humans in making final decisions about who gains access. Like college admissions and SAT/ACT scores, the decision-makers use test scores as one metric among many and use their own discretion to make final decisions. Just as a high SAT or ACT score will not guarantee admission, neither will a high Safety Score guarantee access to FSD Beta. Past driving data and further context (see point #3 below) may illuminate problems that the Safety Score missed.
Misunderstanding #2: If your safety score is low, you’ll never gain access to the $10,000 FSD you paid for -- No!
Sad and/or angry responses to low Safety Scores seem to imply that drivers are being swindled by Tesla. But Safety Scores are calculated daily and there are continual opportunities to raise your Safety Score. A low score this week does not inevitably mean a low score next week.
FSD Beta is only a step towards fully autonomous driving. What people have paid for is FSD itself whenever it is finally released; whether they are eligible to participate in FSD Beta and to help Tesla’s driving computer improve is dependent on the safety of their current driving. But no one is losing access to the final FSD product, even if they may now be denied access to FSD Beta.
Misunderstanding #3: The Safety Score criteria doesn’t pay attention to context, or situations where hard braking or sudden acceleration are necessary to avoid an accident -- Partially true for now.
This is perhaps the biggest criticism of Tesla’s Safety Score. Everyone knows that some situations require hard braking to avoid a collision and it seems unfair to penalize a driver for actually driving safely. And in some situations, the need to brake hard is not the fault of the driver, but of outside factors like other cars or pedestrians.
What critics are forgetting is the sensor suite every Tesla car has: 8 cameras giving a 360 view surrounding the car. And critics are also forgetting all the data labelling Tesla has been doing for the past three years since first explained during Tesla’s AI Day.
Each Tesla car’s eight cameras pick up more “context” than any human ever could. Tesla cars already have the ability to identify green/yellow/red lights, stop signs, animals, and debris on the road. It will be an easy step for Tesla’s team to combine the car’s camera data together with the numerical Safety Score data, in order to gain a more well-rounded understanding of the driver’s situation.
Tesla has the ability to build in context-awareness to actions like hard braking, and we imagine they likely will incorporate it eventually (after all, the Safety Score is also in BETA). Nevertheless, given Tesla's purpose of safely gathering data to do a gradual roll-out to its safest drivers, it's better for them to err on the side of being excessively cautious with these initial Safety Scores
Misunderstanding #4: Being a good driver of a gas-powered car = being a good driver of an EV car.
People who have never driven an electric car misunderstand the nature of braking and acceleration with an EV: (1) An EV has instant torque at a constant rate, meaning that the car accelerates more quickly, but also more smoothly than a gas-powered car.
Drivers accustomed to gas-powered cars will tend to over-accelerate until they get used to an EV’s instant torque, thus negatively affecting their Safety Score.
(2) On the flip side, an EV can engage in regenerative braking. The traditional method of braking uses friction by applying brake pads to the wheels. The friction creates heat that dissipates as “wasted” energy. With an EV, when you take your foot off the pedal, the electric motor stops supplying power, so the car will naturally slow down. When the motor stops, it immediately disengages, and then starts running backwards. The transmission is still in Drive, so it doesn’t reverse the wheels; instead, it acts like a generator. The car captures the kinetic energy from the wheels as they slow down, and converts it into electricity.
This energy capture means that an EV slows down much quicker than a gas-powered car when both are simply “coasting” (no foot on the brakes or accelerator pedal). Many drivers new to an EV instinctively hit the brakes rather than allow regenerative braking to take place.
The term “one-pedal driving” has been used to describe EVs since one could theoretically do most driving simply by using the accelerator pedal to move forward, then letting off the pedal and allowing regenerative braking to slow down the car. Of course, there are situations where immediate braking is needed, which is why EVs still have traditional brake pads that use friction. But most EV driving can be accomplished without the use of traditional braking (which also reduces wear and tear on the car’s brake pads).
Drivers unfamiliar with an EV’s regenerative braking will tend to “hard” brake more often than necessary, thus lowering their Tesla Safety Score.
But with practice, EV drivers can both increase the range on their car and brake more smoothly and safely via regenerative braking.
Misunderstanding #5: Less people gaining access to FSD Beta should be viewed as a good thing.
The main purpose of FSD Beta is to train Tesla’s AI for autonomous driving. Thus, Tesla would naturally only want good drivers to be teaching its AI. No one who purchased FSD will be left out of the final product, but Tesla will weed out bad drivers through this FSD Beta program and that is a good thing in the long run.
Postscript: How to Improve Your Tesla Safety Score
Based on Volt Equity's driving a Tesla Model Y and scoring a 98/100, we've determined that some car settings can be tweaked to help improve your Safety Score:
ACCELERATION: Should be "Chill" mode, not "Standard" more. This helps to slow down turns and decrease following distance.
STEERING MODE: Should be "Standard" mode, not "Sport." This also helps with slowing down turns.
STOPPING MODE: Should be "Hold" mode, not "Creep" or "Roll." This relates to regenerative braking discussed under misunderstanding #4 above. "Hold" mode will apply regenerative braking whenever your foot comes off the accelerator. "Creep" mode will mimic a gas-powered car and still move slowly when your foot comes off the accelerator. "Roll" mode slows down somewhat in between "Creep" and "Hold" modes.
SPEED LIMIT WARNING should be set to Chime (as opposed to Off or only a visual warning).
SPEED LIMIT should be set to "Relative" (to track the speed zone you are driving in), and set to give you a warning when you go over the speed limit by 4mph.
FORWARD COLLISION WARNING should be set to Early. If set to Medium, you will already be penalized by the time of a warning.
These are suggestions, but they will help drivers drive more safely and likely improve their scores.