Thatcham Research, a UK-based agency that works for the Association of British Insurers, has concluded that manufacturers such as Tesla need to be clearer and more specific about the differences between the capabilities of its Autopilot driving-assistance system and full self-driving suites like those employed by Waymo and GM Cruise. In an interview with BBC News, Matthew Avery of Thatcham Research stated that features with names such as Tesla’s Autopilot invoke the impression that the vehicles can drive themselves under all circumstances, when they actually can’t.
“There’s a problem with the manufacturers trying to introduce technology and consumers not being ready for it, not being sure if it’s automated or ‘Do I need to keep watching?’ We want it very clear. Either you are driving – assisted – or you’re not driving – automated,” Avery said.
Avery conducted a demonstration of the risks drivers face if they become over-reliant on the capabilities of the driver-assist system. Engaging a Tesla Model S’ Autopilot, Avery opted to do two tests. In the first test, which involved the electric car following a vehicle that slowed to a halt, the Model S performed well. In the second test, however, which involved a lead vehicle changing lanes without warning, the Model S was not able to stop in time, colliding with a stationary car represented by an inflatable dummy in Thatcham’s test track.
Tesla uses an array of 8 cameras, 12 ultrasonic sensors, and a forward-facing radar to detect objects and other vehicles on the road. Enhanced Autopilot, the company’s driver-assist system that utilizes Tesla Vision, uses a combination of Traffic-Aware Cruise Control (TACC) and Autosteer. Traffic-Aware Cruise Control enables drivers to set a speed and follow a lead vehicle while keeping a fixed distance. Autosteer, on the other hand, detects lane markings and the presence of vehicles and objects, steering Model S based on the road’s markings.
There are some instances, however, where TACC would prove inadequate, such as when the lead vehicle is traveling below the user’s set speed before switching lanes. In such a case, the Tesla would match the lead vehicle’s slower speed to prevent a collision, and once the lead vehicle leaves the lane, the electric car would accelerate to achieve its target speed. Based on the results of the preliminary investigation of the NTSB, this scenario was reflected in the fatal Model X crash in Mountain View, CA earlier this year, with the electric SUV accelerating after the car directly in front departed from the lane.
Such instances, however, only happen when drivers become too reliant on Autopilot. If drivers are paying attention to the road, a simple finger-flick of the Tesla’s stalk to disable TACC is enough to stop the car from accelerating and hitting a stationary vehicle. An alert driver would also be able to apply enough brakes and take over the car’s steering on time to avoid a collision. Tesla actually mentions these in the Owner’s Manual of its vehicles.
“Traffic-Aware Cruise Control does not eliminate the need to watch the road in front of you and to manually apply the brakes when needed. It is your responsibility to stay alert, drive safely, and be in control of the vehicle at all times. Never depend on Traffic-Aware Cruise Control to adequately slow down Model S. Always watch the road in front of you and be prepared to take corrective action at all times. Failure to do so can result in serious injury or death.”
A Tesla Model S crashes into an inflatable dummy. [Credit: Thatcham Research/YouTube]
As noted in a Jalopnik report, Tesla has responded to the results of Thatcham Research’s test, with the company stating that the Model S’ collision would not have happened had the driver been using Autopilot correctly. The California-based electric car maker further noted that Tesla users understand that their attention on the road is paramount when Autopilot is engaged.
“Tesla has always been clear that Autopilot doesn’t make the car impervious to all accidents and the issues described by Thatcham won’t be a problem for drivers using Autopilot correctly. The feedback that we get from our customers shows that they have a very clear understanding of what Autopilot is, how to properly use it, and what features it consists of,” Tesla noted in a statement to the publication.
In order to increase the safety of its drivers, Tesla has been steadily rolling out safety features for its fleet. The latest update to Autopilot, v8.1(2018.21.9), included a more assertive hands-on alert system that ultimately caused complaints from drivers, with some stating that the system’s reminders have become too intrusive. Musk, for his part, has pledged to adjust Autopilot’s screen alerts to attain a balance between safety and drivers’ convenience. If Musk’s recent Twitter updates are any indication, however, it would only be a matter of months before the company starts rolling out the first full self-driving features to its fleet with the release of Software Version 9 in August.
Watch Thatcham Research’s segment on Tesla’s Autopilot in the video below.
Interested in solar? Get a solar cost estimate and find out how much a solar system would cost for your home or business.
Recommended for you
Top Rated Comments (view all)
Discuss: Model S
Join the discussion…
Log in with:
Or sign up with TESLARATI:
1 likes 4 days ago
0 likes 5 days ago
0 likes 4 days ago
Can someone explain why a stationary object is not recognized while a moving object is recognized?
The TACC certainly recognizes the stationary object but above 60mph(highway cruising speeds) the software disregards them as false positives since slamming on the brakes for a phantom objects (like an overhead sign or an aluminum can on the road with the bottom facing the car, can look as big as a car to the radar because of how the radar waves are reflected like a satellite dish) To avoid those false positives the software ignores stationary objects while moving above 60mph. Below 60 mph TAAC *should* recognize and slow for a stationary car. TLDR it doesn’t respond to the stationary objection because it is programmed to ignore it at highway speeds for driver comfort and prevent false braking events. Moving objects are easier for the system to determine if they are real or not because the change in closing distance are changing over a longer period of time for the system to recognize, assess, and act.
0 likes 4 days ago
Currents sensors would brake confused by guard rails, fences, poles, trees and any object around the road.
Neural networks with LiDAR should be smart enough to distighuish real dangers to false positives.
0 likes 4 days ago
0 likes 4 days ago
0 likes 4 days ago
Either fix it or turn it off.
This system will slam your vehicle right into a disabled vehicle if your lead car moves quickly to avoid the same issue.
If the vehicle ahead of you obstructs your view.. you will have no time to react, brake, or otherwise take an evasive maneuver
It will be too late.. Faberge egg cracked.
I have now logged over 28000 accidents free miles since disabling AP
0 likes 4 days ago
0 likes 4 days ago
Current systems ignore static object as they can not distinguish them from thier background.
Currents sensors would brake confused by guard rails, fences, poles, trees and any object around the road.
Neural networks with LiDAR should be smart enough to distighuish real dangers to false positives.
LIDAR does not work very well in fog or rainy conditions so it ultimately won’t be very good for real world driving.
0 likes 4 days ago
0 likes 4 days ago
TESLA SHOULD IMMEDIATELY MOVE TO CORRECT THIS DEADLY ISSUE AND DEMONSTRATED SYSTEM SHORTCOMING.
Either fix it or turn it off.
This system will slam your vehicle right into a disabled vehicle if your lead car moves quickly to avoid the same issue.
If the vehicle ahead of you obstructs your view.. you will have no time to react, brake, or otherwise take an evasive maneuver
It will be too late.. Faberge egg cracked.
I have now logged over 28000 accidents free miles since disabling AP
Note that the exact same is true for a human driver.
A vehicle ‘parked’ in a (left) lane on a highway is almost certain to get hit as are motorcycle drivers thrown of of their bikes.
To avoid such accidents a human driver makes sure he/she/it can view through the vehicle in front, or else increase the distance as much as needed to be able to react on time.
0 likes 3 days ago
The press recently reported Tesla does not have a feature in their Autopilot system that properly monitors if a driver is paying proper attention to the road or able to take over from the AP. While that process cannot actually be made consistently safe, more on the below, this is yet another example of Elon Musk misleading his customers and their families, into paying for the privilege to risk their lives to be Guinea pigs in order to create an autonomous vehicle. The problem is the process Tesla and most AV makers use will never get remotely close to producing a truly autonomous vehicle. Just as with the driver who was banned from driving in the UK for using Tesla’s Autopilot (AP) from the passenger seat a couple weeks ago, this demonstrates Tesla wants its drivers to let go of the steering wheel no matter what their legalese states. If Tesla actually wanted people to keep their hands on the wheel their system wouldn’t allow it to happen. Of course the reason for this is that they need for their customers to not only let go of the wheel but get in the very same accidents they will later scapegoat and abandon them for getting in to. Elon Musk needs to mislead, use and then abandon his paying customers so he can use them and their families as Guinea pigs to train the AI in the “Autopilot” system. The problem is this approach is futile and killing people for no reason.
Public shadow driving is the process where a driver alternates between hands on driving to train the AI and letting go of the wheel so it can be evaluated. Far too many other companies do this as well, Uber being another example. However, Tesla, comma.ai and some others take it further by use their paying customers and their families, not paid drivers, as Guinea pigs. The process is futile because it is impossible to drive and redrive, stumble and restumble on enough scenarios enough times to get anywhere close to L4. Toyota estimates the miles needed are one trillion. My very conservative calculation for cheap cars, sensors and drivers, no other costs, says that is $300B over ten years. To make matters worse though is how unnecessarily dangerous the process is. The process of the vehicle handing over control back to the driver cannot be made consistently safe by any monitoring and control system, especially in complex and dangerous scenarios. This because they cannot provide you the 6-45 seconds of time to gain the proper level of situation awareness to take the right action. Beyond this the AV makers will have to move from public shadow driving hyped benign or simple scenarios to those that are complex and dangerous. Eventually running thousands of accident scenarios thousands of times each. That will mean thousands of casualties.
Unfortunately Elon Musk doesn’t get any of this. He actually said the lives lost in the pursuit of autonomous vehicles are an ends to a means. That some lives need to be lost now to save many more later. The problem being this is not accurate regarding the use of public shadow driving. Musk has also said he only needs 6B miles to get to L4. I believe 4B have been driven so far. That has resulted in Tesla bragging it can now drive straight on a narrow bridge, killing Walter Huang by having his car follow pavement lines vs street marking directly in to a barrier and driving one of their cars into a stopped firetruck. (And that because these systems can’t see non-moving objects right in front of them well. Leaving them to ignore them and drive right through them). On the other side if this though Elon Musk would like to avoid accountability and liability. So Tesla built some legal CYA into their manuals and contracts telling the driver to not let go of the wheel. And if they do they will be to blame for the accidents they are involved in
Elon wants it both ways. He wants to convince his overly trusting customers to pay $5000 for the privilege to risk their lives as Guinea pigs to create a system that will supposedly save lives. Appealing to their desire to be part of a higher good. But he also wants to avoid blame and liability so he creates a legal protection system that abandons his customers when they get in to the accident Tesla needs them to get in. This so the AI can be properly tested. This is why you can see videos of him in his cars with his hands off the wheel. He sends mixed messages to fool people into being his beta testers. All while using a process that will never get close to creating the systems he has convinced his customers and the public will eventually save lives.
What is needed here is for the industry to switch from the use of public shadow driving to the use of aerospace/DoD level simulation. And for someone to step in and force Tesla and Elon Musk to do the right thing. Not unlike what NASA did in 2011 when it said SpaceX code failed safety testing. SpaceX was forced to regroup, hire aerospace engineers and do things right. NASA saved Elon from himself. NHTSA should put a stop to this practice or at least impose a moratorium while these issues are properly vetted. The problem there being that NHTSA is part of the problem. They stated in 2015 that handover can be made safe. The issue there being the study they used to support that position was poorly executed. They determined full control of the vehicle was attained, after being distracted, by simply grabbing the wheel and facing forward. They purposefully chose to ignore the quality of the actions taken after gaining “total control”, the proper situational awareness time needed to perform them properly and if that time could be provided by the monitoring and alert system.
As you can see we have a perfect storm here. The watchdogs are part of the problem. So what is the real solution? Short of more people dying, especially the first child or family needlessly? We need to people, especially those who make up the press and various levels of government, to stopped being overly impressed by the folks who make the apps of their devices or the games that they play, do their due diligence and police an industry that is clearly unable to police itself.
For more information on this please find my other articles
Impediments to Creating an Autonomous Vehicle
• https://www.linkedin.com/pulse/impediments-creating-autonomous-vehicle-michael-dekort/
Autonomous Levels 4 and 5 will never be reached without Simulation vs Public Shadow Driving for AI
• https://www.linkedin.com/pulse/autonomous-levels-4-5-never-reached-without-michael-dekort
DoT, NHTSA and NTSB are Enabling Autonomous Vehicle Tragedies
• https://www.linkedin.com/pulse/dot-nhtsa-ntsb-enabling-autonomous-vehicle-tragedies-michael-dekort/
0 likes 3 days ago
The press recently reported Tesla does not have a feature in their Autopilot system that properly monitors if a driver is paying proper attention to the road or able to take over from the AP. While that process cannot actually be made consistently safe, more on the below, this is yet another example of Elon Musk misleading his customers and their families, into paying for the privilege to risk their lives to be Guinea pigs in order to create an autonomous vehicle. The problem is the process Tesla and most AV makers use will never get remotely close to producing a truly autonomous vehicle. Just as with the driver who was banned from driving in the UK for using Tesla’s Autopilot (AP) from the passenger seat a couple weeks ago, this demonstrates Tesla wants its drivers to let go of the steering wheel no matter what their legalese states. If Tesla actually wanted people to keep their hands on the wheel their system wouldn’t allow it to happen. Of course the reason for this is that they need for their customers to not only let go of the wheel but get in the very same accidents they will later scapegoat and abandon them for getting in to. Elon Musk needs to mislead, use and then abandon his paying customers so he can use them and their families as Guinea pigs to train the AI in the “Autopilot” system. The problem is this approach is futile and killing people for no reason.
Public shadow driving is the process where a driver alternates between hands on driving to train the AI and letting go of the wheel so it can be evaluated. Far too many other companies do this as well, Uber being another example. However, Tesla, comma.ai and some others take it further by use their paying customers and their families, not paid drivers, as Guinea pigs. The process is futile because it is impossible to drive and redrive, stumble and restumble on enough scenarios enough times to get anywhere close to L4. Toyota estimates the miles needed are one trillion. My very conservative calculation for cheap cars, sensors and drivers, no other costs, says that is $300B over ten years. To make matters worse though is how unnecessarily dangerous the process is. The process of the vehicle handing over control back to the driver cannot be made consistently safe by any monitoring and control system, especially in complex and dangerous scenarios. This because they cannot provide you the 6-45 seconds of time to gain the proper level of situation awareness to take the right action. Beyond this the AV makers will have to move from public shadow driving hyped benign or simple scenarios to those that are complex and dangerous. Eventually running thousands of accident scenarios thousands of times each. That will mean thousands of casualties.
More here
Elon Musk is the Pied Piper of Autonomous Vehicles
https://www.linkedin.com/pulse/elon-musk-pied-piper-autonomous-vehicles-michael-dekort/
0 likes 2 days ago
LIDAR does not work very well in fog or rainy conditions so it ultimately won’t be very good for real world driving.
Correct.
The only possibility is having well trained neural networks.
I would like to know if Waymo has already solved this issue. They are talking about an introduction of self driving taxis by this year, they must have solved this.
0 likes 2 days ago
Tesla hit parked police car ‹while using Autopilot
www.bbc.com/news/technology-44300952
And
[MEDIA=twitter]1001541486146547717[/MEDIA]
0 likes 2 days ago
TESLA
Tesla on fire: UK director Michael Morris› car bursts into flames
4 hours ago
0 likes 1 day ago
[MEDIA=twitter]1008164785844715520[/MEDIA]
