Disqus - Latest Comments for imispgh

Re: A designated driver for the driverless car

Michael DeKort — Sat, 17 Feb 2018 22:28:35 -0000

While this use case may be necessary and there are benign slow speed situations where the setup I see Phantom Auto using will work. What I see is a non-full motion DIL. And what I assume is a cloud connection. I am concerned about round trip latency and the driver not getting motion cues. Both of these can lead to extremely poor operability. Especially where the speed is high, there are quick movements, on grades and where traction is not good. (You state you use multiple networks cellular networks etc. I do not believe the data throughput round trip is fast enough. This is not gaming. Something not evident in videos or at slow speed etc. What is the latency from remote wheel move to target wheel move? Now from that action to the remote driver seeing the wheel move and the change? How often does data need to be retransmitted? 2nd, 3rd packets etc?)

How easy is it to hack this and weapon that vehicle? If Phantom Auto uses even the bes of the normal cybersecurity best practiced used out there is is an easy target.

Can someone alleviate my concerns?

Re: California's latest self-driving report signals long road ahead

Michael DeKort — Wed, 31 Jan 2018 21:23:18 -0000

I am all for gathering data and training AI while driving. My problem is with letting go of the wheel and testing AI. That is from a technical POV. From a time, money, safety and the ability to be successful in actually creating an autonomous system you have to switch to simulation. And not the sim systems this industry uses and most sell. The problem is it isn't being done right on several levels. You have to integrate AV sensor simulation with a full motion Driver-in-the Loop simulator, in proper real-time and create scenarios from the top down using a massive cadre of several different domain experts, aerospace type systems engineering and data from sources other than folks driving around stumbling on things.

As for the firetruck. My guess is the sensors saw the truck The logic used with the data was wrong.

Re: California's latest self-driving report signals long road ahead

Michael DeKort — Wed, 31 Jan 2018 19:23:15 -0000

Folks disengagements and miles driven mean almost nothing without scenario and root cause data. Please do the homework. All that matters is scenarios learned, how many are left and what that whole set is. Ever sen that? No one has and there is a reason.

Public Shadow Driving is Dangerous and Untenable.
The reason it is not possible to create an autonomous vehicle using that process is that it will take one trillion miles at a cost of over $300B. And those using it will cause thousands of accidents, injuries and casualties when they move from benign scenarios to complex, dangerous and accident scenarios.

See my Linkd article - Autonomous Levels 4 and 5 will never be reached without Simulation vs Public Shadow Driving for AI

Re: Perhaps "Stereo Stores" Are Designed All Wrong - Audiophile Review

Michael DeKort — Sat, 09 Sep 2017 11:17:56 -0000

You can have the minority come to you or you go to them.

Seems to me that a larger trailer in the right location or a store in the mall that the entire industry supports to educate people would work best. Have a low and mid price system and headphones. Let folks use their devices and music then you do the same. Also let them hear quality differences by bit rate etc.

An nonprofit association could handle the effort. Have folks pay dues. As for equipment it will be a food fight so I suggest hiding it. Then provide a buyer's guide and let folks do their thing. You sell NOTHING.

Re: Mobileye, Intel plan to build 100 Level 4 autonomous test cars

Michael DeKort — Mon, 14 Aug 2017 10:20:25 -0000

Lockheed Engineer/Whistleblower

It is very unfortunate and counter-productive for Intel to take the approach they are.

Autonomous Levels 4 and 5 will never be reached without Simulation vs Public Shadow Driving for AI.

Level 2+and L3 should not be used at all.

Public Shadow Driving is Dangerous. Thousands of accidents, injuries and casualties will occur when these companies move from benign and easy scenarios to complex, dangerous and accident scenarios. And the cost in time and funding is untenable. One trillion public shadow driving miles would need to be driven at a cost of over $300B. The answer is aerospace level simulation.

For details please search for my LinkedIn article

Autonomous Levels 4 and 5 will never be reached without Simulation vs Public Shadow Driving for AI

Re: Waymo Now Inviting Real People To Test Its Self-Driving Minivans

Michael DeKort — Thu, 27 Apr 2017 14:25:29 -0000

Lockheed engineer and whistleblower

Waymo – What, ethical, moral and technical reasons do you have to make children Guinea pigs?

Waymo is now signing up families to be beta test subjects in their autopilot vehicles. Vehicles that are not nearly complete with data collection, design or testing. Based on the questions/statements below I would like to know what ethical, moral or technical value does this have? How is this doing anything other than providing the public with false confidence and putting them needlessly at risk?
• NHTSA has deferred to industry to determine best practices. In doing so they avoid their due diligence in creating minimal test standards to ensure the proper scenarios are covered, covered the same vendor to vendor and that sensors are integrated properly. (That being with redundancy and tested to make sure they always give the best answer.) This includes accident scenarios.
o Is Waymo, or anyone else who is doing things like this, showing those passengers or anyone else that Scenario Matrix? Especially negative testing results?
• MIT has shown that AI, machine learning, neural networks etc have flaws. Especially regarding outliers or corners cases. Many of these will be accident scenarios or cause one. No one is remotely close to collecting that amount of data.
• RAND has shown that these companies massively underestimate the amount of miles that need to be driven to gather the data needed and prove these vehicles are as good and progressively better drivers than people. They say it would take 11 billion miles of driving to prove they are only 20% better than people. That extrapolates to hundreds of billions of miles to prove they are 2X as good and far, far more to prove the 10X industry has said they need. (And exactly how are you collecting the critical accident scenario data? Driving around with people on board waiting to stumble on them so you can have the data?)
• Ford as well as studies by the EU shown that professional drivers who are monitoring AP vehicles with their hands off the wheel cannot stay attentive. Many even falling asleep. Other studies have shown people cannot react fast enough in many situations and even over react when they do. Think about it. It that vehicle moved quickly to either side the driver’s body is going to react. That will delay time to grab the wheel. And they are already limited by not paying the same attention as a driver so their response will be delayed due to a lack of data they need to make a decision on what to do.
• Why do you need human Guinea pigs in early AP mode at all? Why does anyone have to let go of the wheel in an AP vehicle? Why can’t you use simulation, other data inputs, test tracks, hands on driving and proper systems engineering practices to do this?

I provide more on this and references for the items I mentioned above in this article - Lockheed Engineer/Whistleblower - Due Diligence Recommendations for Autonomous and Driverless Industry -https://www.linkedin.com/pu...

Re: Cyber maturity will help ensure cybersecurity

Michael DeKort — Wed, 19 Apr 2017 14:28:40 -0000

Privileged Account Security – The Giant Dirty Secret and massive hole in most organizations cybersecurity. Why isn't it being addressed? Lack of Courage.

The overwhelming majority of companies and government organizations are avoiding the most critical cyber-security practice of all. Dealing with privileged account security. It’s the biggest dirty secret in cybersecurity. Which is extremely unfortunate because virtually every hack on record was accomplished by someone gaining access to a privileged account then moving through the system. This usually occurs due to a successful phishing expedition. (Of which 22% are successful. Keep in mind only one is needed).

Of the small fraction of companies that even deal with this area only 1% of them actually use the products they purchase properly. Said differently – even if a CISO is buying the right things they are not using most of what you paid for. And in most cases they either have no plan to actually use critical features like Password Management, Session Management and Access Monitoring, or are moving so slow it will decades to finish. This puts everyone at risk.

Here is how bad things are. CMU CERT is the premier authority on cyber-security best practices. Especially for DoD. I found out that CMU CERT has no solution for themselves in this area. They actually defer to CMU IT for their own security and they have no solution in this area. Shouldn't the organization responsible for telling others what best practice is use best practices for its own security?

Why is this happening? These products inadvertently expose several huge best practice gaps. Examples include having 4X more accounts than people, non-encrypted password files or spreadsheets, emails with passwords and software programs with passwords hard coded in them and many not knowing where they all are. Why don't these folks address this? Because it means pushing the culture to change bad habits and admit to their executives and boards they even existed in the first place. Governing bodies and regulators mean well but they don’t help much. This is because the relevant regulations, SOC, HiTrust etc are too trusting and don’t specify enough detail. This gives organizations far too much room to wiggle. This all results in most companies and organizations not utilizing best practices or readily available of off the shelf products that can significantly reduce the threat. Or when they do purchase these products they only use a small fraction of their capabilities.

This is not a technical issue. It's one of Courage. Courage to admit the root causes exist, To deal with the culture and lead them to fix them. To not sacrifice customers to protect egos or let the bean counters justify it's cheaper to harm customers than the bottom line.

Re: Tesla made it difficult for hacker geohot to buy a Model S and use his own software to power Autopilot 2.0

Michael DeKort — Tue, 07 Mar 2017 10:22:36 -0000

I am going to support Tesla on this one. For several reasons
- Comma.ai releasing the code was extremely reckless. Very few if anyone who owns those systems, especially Commercial IT devs have enough experience to do this. Especially those who think they do because they made cool apps, websites or games. Without KNOWING how all of the code impacts all of the scenario threads changing any of this code is extremely dangerous. (Especially since that scenario matrix is probably barely complete itself). The other reason is that these cars can easily be weaponized.
- Tesla wants no part in that recklessness especially where it would regard public opinion or legal issues when, and I mean when, comma, ai gets people killed.
- Tesla has no reason to go out of its way to help a competitor. (Clearly if Hotz wants the car someone else can buy it for him)

NHTSA and DHS should stop comma.ai's reckless behavior and shut them down.

Re: Tesla Autopilot Sends Model S Into Concrete Barrier

Michael DeKort — Sun, 05 Mar 2017 21:22:44 -0000

Saying you work in Lockheed as a systems engineer with any kind of experience means you usually have experience in not just systems engineering best practices but probably to a CMMI Level 5? That would often include exception handling. Things most folks in Commercial IT have little experience in.

Maybe you could discuss my points one by one, telling me why I am incorrect and why for each one, to ensure it is obvious to everyone I know absolutely nothing?

Re: Tesla Autopilot crash caught on dashcam shows how to not to use the system

Michael DeKort — Sun, 05 Mar 2017 13:41:36 -0000

I have updated the info above with references. (And I added one for lack of best practices industry wide.)

I guess we will let the audience determine if there are no primary
sources. BTW - how do you KNOW I. or the references I posted, are wrong? (And please provide references of your own).

And it would be helpful to know what your background is and who you have worked for.

Re: Tesla Autopilot crash caught on dashcam shows how to not to use the system

Michael DeKort — Sun, 05 Mar 2017 11:39:12 -0000

It is unfortunate you are resorting to shooting the messenger and avoiding the discussion of the issue objectively. I posted it elsewhere but will again here. And while doing so without having the courage or ethical fortitude to explain your background. Please respond as to exactly why the method I describe below is not preferable/

Suggested Method for Data Capture, Design and Testing - Autopilot use in the public would be based on a proven progressive set of test data from this process

Scenario Matrix

- The key to all of this is creating a complete and accurate scenario matrix which would be used for design and testing. (That includes regression testing)

- That matrix should include any situation that a user of the system could reasonably be expected to experience

- Design and Testing should include the boundaries of those combinations.

Data Sources

- Drivers not in autonomous mode

- Automobile Companies

- Insurance Companies

- Researchers – to include Social Engineering –
Expectations of other human drivers

- Traffic Engineering

- Government Agencies – NHTSA etc

- Product Team Exception Handling Inputs – Folks trying to break the system

- Vehicle Performance, Configuration Changes and Issues

- Weather

- Road and Terrain – Time of Day

- Signage - To include human gestures

- Sensors – System Device Capabilities - Handling of Conflicts and Missing or Flawed Data - Priority and Probability Filters - LIDAR, Radar, FLIR, GPS, Cameras, Sound etc

- Moving and Stationary Objects

- External Data Sources – Other Vehicles, Objects and Systems

- Route deviations based on interior and exterior changes - Include handling of latency

- User Error – When to ignore, correct or notify

- System Wide Conflicts, Missing Data or Errors

Use of Data, Design and Testing Approach

- Create the Scenario Matrix in an Object Oriented software system that represents the combination of all the various data and system types. Once the data areas and exception boundaries are know the various combinations of them can be created, tuned, changed and tested.

- Utilize non-manned and manned simulations/simulators to run through the various scenarios.

- Utilize real world testing from test tracks and controlled public driving to verify the simulations.

Re: Tesla Autopilot crash caught on dashcam shows how to not to use the system

Michael DeKort — Sun, 05 Mar 2017 11:31:48 -0000

Suggested Method for Data Capture, Design and Testing

Scenario Matrix

The key to all of this is creating a complete and accurate scenario matrix which would be used for design and testing. (That includes regression testing)

That matrix should include any situation that a user of the system could reasonably be expected to experience

Design and Testing should include the boundaries of those combinations.

Data Sources

Drivers not in autonomous mode

Automobile Companies

Insurance Companies

Researchers – to include Social Engineering – Expectations of other human drivers

Traffic Engineering

Government Agencies – NHTSA etc

Product Team Exception Handling Inputs – Folks trying to break the system

Vehicle Performance, Configuration Changes and Issues

Weather

Road and Terrain – Time of Day

Signage - To include human gestures

Sensors – System Device Capabilities - Handling of Conflicts and Missing or Flawed Data - Priority and Probability Filters - LIDAR, Radar, FLIR, GPS, Cameras, Sound etc

Moving and Stationary Objects

External Data Sources – Other Vehicles, Objects and Systems

Route deviations based on interior and exterior changes – Include handling of latency

User Error – When to ignore, correct or notify

System Wide Conflicts, Missing Data or Errors

Use of Data, Design and Testing Approach

Create the Scenario Matrix in an Object Oriented software system that represents the combination of all the various data and system types. Once the data areas and exception boundaries are know the various combinations of them can be created, tuned, changed and tested.

Utilize non-manned and manned simulations/simulators to run through the various scenarios.

Utilize real world testing from test tracks and controlled public driving to verify the simulations.

Re: Tesla Autopilot crash caught on dashcam shows how to not to use the system

Michael DeKort — Sun, 05 Mar 2017 10:24:51 -0000

You are dodging the data.

What is your background? What companies have you worked for?

How is using human Guinea pigs safer and faster than
simulation/simulators with the teams and process I describe in my
LinkedIn article?

Please find someone who has actual experience in systems engineering especially around exception handling and simulation to discuss this. Someone who takes the subject seriously and who won't avoid an in the weeds discussion.

Re: Tesla Autopilot Sends Model S Into Concrete Barrier

Michael DeKort — Sat, 04 Mar 2017 23:33:51 -0000

Lockheed Engineer/Whistleblower

Tesla's Autopilot needs to be shut down and NHTSA needs to do their due diligence

https://www.linkedin.com/pu...

Musk doesn't keep his hands on the wheel - https://www.youtube.com/wat...

Re: Tesla Autopilot crash caught on dashcam shows how to not to use the system

Michael DeKort — Fri, 03 Mar 2017 13:16:06 -0000

You are clearly dodging my points and playing game with an important subject.

Anecdote - an account regarded as unreliable or hearsay

Data - facts and statistics collected together for reference or analysis

So now tell me which parts of what I have said is in either category. Especially my suggested method for doing this safer and faster. And if you don't think my suggested plan for doing so makes sense please tell me exactly why.

Re: Tesla Autopilot crash caught on dashcam shows how to not to use the system

Michael DeKort — Fri, 03 Mar 2017 12:44:27 -0000

The NHTSA finding, which I disagree with, was based on the previous version of software that didn't have the car steering itself across lanes. A clear regression of capability.

Second - the NHTSA finding did not concern itself with the method Tesla uses to collect data, design and test it's systems. And it certainly did not review the alternative I suggested.

So again - is that recent software release a problem? Are you maintaining it meets the 40% less accident criteria of the previous release? And more importantly EXACTLY why is the current method superior to the one I suggested or in the reverse why isn't my recommendation safer and faster?

Re: Tesla Autopilot crash caught on dashcam shows how to not to use the system

Michael DeKort — Fri, 03 Mar 2017 10:23:18 -0000

Time will tell

Why is Elon doing exactly what the small print says not to do here - https://www.youtube.com/wat...

Re: Tesla Autopilot crash caught on dashcam shows how to not to use the system

Michael DeKort — Fri, 03 Mar 2017 10:18:53 -0000

This is what the upgraded version does -

Tesla's #Autopilot needs to be shut down and #NHTSA needs
to do their due diligence - https://www.linkedin.com/pu...

Regardless of version there is a much safer and faster way to do this. People need not be autopilot Guinea pigs. Whether in the vehicles in the area of one.

Re: Tesla Autopilot crash caught on dashcam shows how to not to use the system

Michael DeKort — Fri, 03 Mar 2017 10:13:52 -0000

My experience in Commercial IT and DoD is not anecdotal. It is real. Do you have the same experience and can tell me I am wrong and why point by point? Instead of hyperbole please tell me why my suggested method for doing this is not safer and faster.

Also please tell me how this new version of the product is safer than someone driving themselves - http://www.thedrive.com/new...

Re: Tesla Autopilot crash caught on dashcam shows how to not to use the system

Michael DeKort — Fri, 03 Mar 2017 10:10:05 -0000

Why didn't you call it an Autopilot?

Re: Tesla Autopilot crash caught on dashcam shows how to not to use the system

Michael DeKort — Thu, 02 Mar 2017 22:27:42 -0000

The official determination on speed has not been given yet. Yes BAC was way over the limit.

As for my points please refer to what my actual points or, more specifically, what my questions are.

https://www.linkedin.com/pu...

Tesla and all of these companies, like in airplane crashes, should supply data when there is an accident especially when there is a death.

Re: Tesla Autopilot crash caught on dashcam shows how to not to use the system

Michael DeKort — Thu, 02 Mar 2017 22:03:22 -0000

Where is the data on this event if there is no there - there? Tesla has no problem trotting data out showing step by step actions when drivers are at fault. This episode went on for a while and there was no accident. There is a clear pattern. When the driver is at fault and Tesla bears none we get immediate play by play. when it's the opposite there is silence or Tesla says there isn't any data.

Where is the data from the drunk Tesla video? No crash and it went on for a while so there should be plenty.
https://jalopnik.com/watch-...

The McCarthy/Speckman tragedy is another example. In that case they said the car was in such bad shape none could be transmitted. Really? None from that day at all? And what good is that system if you miss crucial data on accidents like that? You can't have it both ways. Do one or the other properly. Either have a system that transmits all data, especially in crash scenarios, have a legit black box or stop using people as Guinea pigs.

Re: CONFIRMED: Obama Sabotaged Trump’s Transition To The White House

Michael DeKort — Thu, 02 Mar 2017 21:35:00 -0000

All they had to do was leak all of it to the press and Trump would have been done. Instead they planted sees so it wouldn't be buried. That's coming from an independent.

Here's how you can tell if your a useless partisan tool. If you have an issue with something the other side does but justify it when your side does it - you are an unethical hypocritical party tool.

Party tools on both sides are a problem. The ridiculous belief that party is more important or equal to being a good American . That somehow your party is 100% right on every issue. You party tools are the problem. What we need are mature, ethical adults who deal with things issue by issue. Part by part. Not blind allegiance to parties.

Re: Tesla Autopilot crash caught on dashcam shows how to not to use the system

Michael DeKort — Thu, 02 Mar 2017 20:52:05 -0000

Then don't call it Autonomous. And expecting a driver to figure out which parts of macro functions the car won't do is ridiculous. Work on on highways or don't. Work on city streets or don't. Using people as human Guinea pigs to figure out how to handle things like this is wrong. And completely.How many of you folks are under 25 and don't have kids? Of those who do would you let your child use autopilot? It's a roll of the dice.

Re: Tesla Autopilot crash caught on dashcam shows how to not to use the system

Michael DeKort — Thu, 02 Mar 2017 20:43:39 -0000

One of the problems I have with this is the levels especially the situation capabilities. Drivers should not have to keep track of this. When it is a clearly separate activity like parking or can help you do something you missed or should have done better - fine. But it's absolutely ridiculous and I believe dangerous fro the driver and the public for autopilots to be engaged or disengaged based parts of a major function. Usually because the vendor hasn't included it yet. Like getting off a highway? Really? Either provide all of the parts of macro functionality or don't.