Tesla Model 3 Autopilot Versus Bikers & Pedestrians


One of the most telling videos to date of Tesla Model 3 Autopilot in busy downtown and residential areas.

First of all, we’ll say that Tesla Autopilot is not meant to deal with busy city streets or neighborhoods. The Autosteer system is still in Beta mode and the technology’s primary intention is to assist drivers on the highway. Additionally, it’s important to read Tesla’s disclaimers about the system and assure that you are using it as advised, keeping your hands on the wheel, and remaining engaged at all times. With that being said, as we think forward to Full Self-Driving capability, it’s interesting to see how the current Autopilot tech in this Tesla Model 3 handles pedestrians, cyclists, and even another car parallel parking. How will the Tesla fully autonomous system deal with such difficult situations?

YouTuber Nick from Nick’s Tesla Life makes many telling observations as he safely navigates these congested areas in his Tesla Model 3 with Autopilot engaged. He notices that at these slower speeds, it may not stop for pedestrians crossing the street, however, there’s no way to know for sure since he must apply the brakes to be cautious. He also learns that it will detect a moving cyclist and treat it as if it is a car, not passing it, but rather following and adjusting speed. The most interesting test is when another car stops and then proceeds to perform a parallel parking maneuver. Watch to find out exactly how this particular Tesla Model 3 Autopilot handles the challenge.

Video Description via Nick’s Tesla Life on YouTube:

Tesla Model 3 Autopilot on a Local Road with Lots of People and Bicyclers

Day 168 – I drive down a busy local road in Ocean City, NJ with a lot of people and bikers on the street in my Tesla Model 3 using Autopilot. About a year ago I drove down a similar street in the same town in my Model S – https://youtu.be/qUwZ_XZVRW0

Current software version: v8.1 (2018.32.2)


33 photos
2. Tesla Model 3 Range: 310 miles; 136/123 mpg-e. Still maintaining a long waiting list as production ramps up slowly, the new compact Tesla Model 3 sedan is a smaller and cheaper, but no less stylish, alternative, to the fledgling automaker’s popular Model S. This estimate is for a Model 3 with the “optional” (at $9,000) long-range battery, which is as of this writing still the only configuration available. The standard battery, which is expected to become available later in 2018, is estimated to run for 220 miles on a charge. Tesla Model 3 charge port (U.S.) Tesla Model 3 front seats Tesla Model 3 at Atascadero, CA Supercharging station (via Mark F!) Tesla Model 3 Tesla Model 3 The Tesla Model 3 is not hiding anymore! Tesla Model 3 (Image Credit: Tom Moloughney/InsideEVs) Tesla Model 3 Inside the Tesla Model 3 Tesla Model 3 rear seats Tesla Model 3 Road Trip arrives in Tallahassee Tesla Model 3 charges in Tallahassee, trunk open.


Tesla Model 3 Performance - Dual Motor Badge
10 photos
Tesla Model 3 Performance Tesla Model 3 Performance Tesla Model 3 Performance Tesla Model 3 Performance - Midnight Silver Tarmac Motion (wallpaper 2,560x – click to enlarge) Tesla Model 3 Performance - White Interior - Wide Tesla Model 3 Performance - White Interior - Touchscreen

Categories: Tesla

Tags: , ,

Leave a Reply

25 Comments on "Tesla Model 3 Autopilot Versus Bikers & Pedestrians"

newest oldest most voted

This is a good illustration of how challenging full self driving will actually be. I would curious how it handles a cyclist in a bike lane.

It has no idea of size and identified the cyclist as a car by default. It therefor would not have allowed passing.

In this video the cyclist wasn’t in an actual marked bike lane where there would be a solid painted line to the left of the cyclist.

Obviously a long way to go. Doesn’t recognizing anything not moving, and he notes that it was driving too close to the parked cars because it didn’t know they were there. Missed most of the pedestrians and was late in identifying cross traffic.

Yes, a long way to go. But no as to not identifying anything not moving, at least according to this video. It would have taken out several parked cars if it “doesn’t recognize anything not moving…”, as you noted according to driving too close to parked cars. It was not late in identifying cross traffic; it simply slows more and takes longer to start up again than normal drivers.

I personally think self driving cars should mostly be used on limited access highways with no stoplights or risk of cross traffic.

The idea of doing this in a busy city street is a liability factory waiting to open.

Here is a anti Tesla and solar power posted by Forbes https://www.forbes.com/sites/judeclemente/2018/09/09/tesla-and-electric-cars-dont-replace-oil-and-natural-gas/#a8a70a737cdc

But according to Elon, AP has been about ten times better at driving than people since ca 2015…

Say what? Elon said Tesla’s goal is achieving a 10x safer than human autonomous driving system. He absolutely never claimed or hinted that this has been achieved! …at least, not yet.

Full self-driving (true level 5) will not work until there is both V2V (vehicle to vehicle) and a dedicated network (not the open internet) all in constant communications plus an enormous increase in onboard computing power to handle all that A.I. data in faster than real time.

There will always be something not connected to the dedicated network like pedestrians and cyclists so self-driving cannot depend on this. Must be able to observe all objects, determine size, distance and if moving how fast and in what direction. I believe the best way to accomplish this is to use binocular vision (pairs of cameras) thus allowing triangulation to determine distance.

For instance, connected street and building cams can feed pedestrian info to the network so there’s much more data for triangular, speed, direction, etc. Many (most?) pedestrians are connected already via their phones.

Tesla Model 2?

That’s a new one

Foolish, dangerous, and irresponsible.

I wonder what would have happened if a police officer came up behind ole Rick while he was putting the public’s lives at risk for his YT fame?

Autopilot limitations
Traffic-Aware Cruise Control is a beta feature.
Traffic-Aware Cruise Control is primarily intended for driving on dry, straight roads, such as highways and freeways. It should not be used on city streets.

Warning: Do not use Traffic-Aware Cruise Control on city streets or on roads where traffic conditions are constantly changing.

Warning: Do not use Traffic-Aware Cruise Control on winding roads with sharp curves, on icy or slippery road surfaces, or when weather conditions (such as heavy rain, snow, fog, etc.) make it inappropriate to drive at a consistent speed.

Auto steer is a beta product.

Autosteer is intended for use only on highways and limited-access roads with a fully attentive driver. When using Autosteer, hold the steering wheel and be mindful of road conditions and surrounding tra c. Do not use Autosteer on city streets, in construction zones, or in areas where bicyclists or pedestrians may be present.

At $5,000 its too much to pay for the limitations.

Then don’t get it.

“Beta” sounds like a product will be released soon. It couldn’t be further from truth.

There is no way we will see self driving cars before 2030, and even that assumes a series of breakthroughs in AI technology. CNNs are a good start but they are completely inadequate for open-ended applications like understanding the surroundings.

Even much simpler problems, like automatic language translation, remain unsolved. Why would anyone expect autonomous driving before good quality translation?

Autosteer is what is in Beta, not Full Self Driving. FSD is still in internal testing, it is not available to anyone outside Tesla. I think Autosteer may come out as a full product in the next year, after V.9 and the hardware swapout probably, but it will be geographically limited to restricted access roads, and you will still be driving at your own risk on other roads. (If not, how is autosteer different from FSD?)

“Even much simpler problems, like automatic language translation, remain unsolved. Why would anyone expect autonomous driving before good quality translation?” Accurate language translation is an extremely challenging problem for “expert systems” software. A human translator can actually understand, on a human level, what information is being communicated. The computer has no such understanding; none at all. All it can rely on is what it’s been pre-programmed for, or at best use some sort of algorithm to find relationships between words in a database of information, which is how IBM’s Watson supercomputer can win at Jeopardy. Contrariwise, all a self-driving system really needs to do is to avoid colliding with any object of substantial size, obey posted traffic signs, and navigate from point A to point B — the last of which can already be done reliably by several different GPS based in-car navigation systems. Autonomous driving isn’t a simple problem, but it’s certainly possible with today’s computer hardware. In fact, if my understanding is correct, Waymo/Google already has cars capable of fully self-driving; they don’t even have steering wheels! However, they are limited to driving at 25 MPH or less, so clearly the system needs upgrade to let the software… Read more »

The car knows it’s in a city because GPS. Why would AP be available at all, when Tesla says “do not use” in this circumstance?

It’ll be interesting to see this route being drivin again in a month or two, after the upcoming autopilot update.

Inappropriate use of EAP. It is for highway driving, not congested city streets. The results are meaningless.

I use ELR’s ACC all the time in my mixed commute. It is especially handy in school zones to maintain max 20mph speed. Yeah, there are limitations, but the speed control is flawless. Where it gets into trouble is detecting cars entering or leaving the lane. Over three years using ACC every day.

AP still doesn’t recognize anything not moving. To me, this merely underscores just how limited these Level 2 (with possibly some limited aspects of Level 3) semi-autonomous driving systems are.

The current goal should be developing a SLAM* system, which would map the entire environment around the car — not just moving objects! For truly safe autonomous driving, the car needs to “know” where the stationary obstacles are, too.

So far as I know, Waymo/Google is the only company which has developed a SLAM for their test cars, and even there it was reportedly limited to use at 25 MPH or less. There is still a long way to go to get to Level 4/5 autonomy. Why are auto makers wasting time fiddling around with minor improvements to Level 2+ systems? Wouldn’t they be better off spending their time developing a SLAM, which has to be the basis of any workable Level 4+ autonomous driving system?

*SLAM stands for Simultaneous Localization And Mapping technology, a process whereby a robot or a device can create a 3D map of its surroundings, and orient itself properly within this map in real time.