I just saw an announcement of a RaspberryPi team winning an autonomous car race. That sounded exciting, but then I read that the following steps were required.
1. Blue and Yellow cones are placed around the track.
2. Humans then choose the optimal path
3. Humans created a map.
Car drives 'autonomously'
I don't get it, I am quite certain Tesla cars do not need coloured cones or humans to pre-map their route.
What the heck is different between what Tesla et al do, these contests (I have seen a few recently) and what y'all are doing with this genetics approach?
Is your goal to create an autonomous car/robot? If so, don't you need to find out what Tesla is doing?
Confused am I.
Although the subject matter has been of great interest to me for decades, the reason i have responded, I didn't want to hijack the thread which is about the progress being made in a particular experiment.
With regards your reference I assume you mean this,
I think the experiment in randell's thread is seeing if he can evolve a set of weights using genetic algorithms for a neural net that will produce a behavior (obstacle avoidance in this case) given a set of inputs (distance measuring sensors) and a set of outputs (two motors).
Self driving cars on the other hand rely on accurate GPS readings and internal road maps just like your GPS navigation system but instead of telling you to turn at 800m the car does the turning itself. An important addition is obstacle avoidance for things not present in this internal map like people and other cars and traffic signs. To detect people and other cars (or any other obstacle) they might use LIDAR, RADAR and visual recognition of traffic lights, road signs as input to the self driving car programs.
The programs use trained (not evolved) neural networks to recognize things like signs or people but the actual programs are not neural networks. They are not evolved solutions either.
@robotbuilder I didn't even consider that they would be using GPS. Either they have a much better GPS than I do, or they heavily augment it. Where I live now, a major highway was modified maybe 20 years ago but the GPS data was not. The same is true for the newest highway in my region. The GPS is wrong enough to get you killed.
I thought they were using similar tech that is in my car. It is NOT self-driving, but if I turn on the right features and just lightly touch the wheel it will do great on that highway I just mentioned (if I don't touch the wheel it nags me and I don't know what else it will do after that). It does rely on road markings though so that is why it is not self-driving. Another system would be needed to augment the vision system I have. I don't know what that is, it may be GPS, but it just may be a better vision system than what I have in my little Ford Edge.
I just googled Tesla self-driving (various search arguments) and I see it uses Google Maps, but also MapBox for Navigation. The main difference is the 8 cameras plus some smart code to keep you from crashing or killing someone.
First computer 1959. Retired from my own computer company 2004.
Hardware - Expert in 1401, and 360, fairly knowledge in PC plus numerous MPU's and MCU's
Major Languages - Machine language, 360 Macro Assembler, Intel Assembler, PL/I and PL1, Pascal, Basic, C plus numerous job control and scripting languages.
My personal scorecard is now 1 PC hardware fix (circa 1982), 1 open source fix (at age 82), and 2 zero day bugs in a major OS.
GPS can be very accurate under the right conditions. It is used extensively in tractors used in broad acre farming.
https://www.techbriefs.com/component/content/article/28021-gps-correction-technology-lets-tractors-drive-themselves
"For more than a decade, John Deere’s StarFire GPS receivers used NASA’s global network of ground stations and the JPL software, which the company licensed, to enable self-guided tractors."
A self driving car that can find its way around (navigate) does need to know where it is and I don't see any other way being used other than GPS to determine that?
In my old i30 the GPS seems to have a good idea of where it is showing a little car indicator on a map as I drive around. I imagine you could use vision to detect say white lines or the side of roads and radar to detect obstacles for more accurate adjustments. I was using a curtesy car while my old i30 car was being serviced and was surprised when the speed control turned itself off. I quickly realized I was too close to a car in front and the car was using radar to release the accelerator and apply the brakes.
Modern cars are incorporating driver support systems as you apparently have in your Ford Edge.
https://www.caranddriver.com/features/g27612164/car-safety-features/
None of them use a 100% evolved neural network as a complete solution.
There are plenty of examples on the internet about genetic algorithms and what they can do.
@robotbuilder I am quite aware of the farming use of GPS, but my personal experience involves a major highway exit to another major highway and the GPS did not allow a left turn, rather it sent you right then pulled a U turn. The fact is the interchange used to be like that but had been reworked to be 'normal' decades ago.
Similarly a new major highway (divided 4 lanes) is simply missing at some points.
A long time ago I experimented and found that the street that is the world's longest street that runs from downtown Toronto all the way to the MB border (1,877Km) had the name spelled several different ways. My research told me that a lot of the GPS data is manually input overseas and is subject to many typos and severe errors. While spelling errors are not harmful, the missing highway is.
First computer 1959. Retired from my own computer company 2004.
Hardware - Expert in 1401, and 360, fairly knowledge in PC plus numerous MPU's and MCU's
Major Languages - Machine language, 360 Macro Assembler, Intel Assembler, PL/I and PL1, Pascal, Basic, C plus numerous job control and scripting languages.
My personal scorecard is now 1 PC hardware fix (circa 1982), 1 open source fix (at age 82), and 2 zero day bugs in a major OS.
While spelling errors are not harmful, the missing highway is.
Yes I understand that but as far as I know they do use GPS to locate the global position of a car for the purpose of computing a path to some global goal position. If the map (GPS data as you called it) is wrong then so will be the computed path.
We kind of got off track on your question as to the difference between Tesla self driving cars and what randell and inq were doing which was evolving a brain. The problem with genetic evolution is you would need a lot (thousands) of Tesla cars with random brains competing for survival and crashing everywhere that is why it is probably simpler to evolve a brain in software using simulated cars and then hope the simulated cars (and simulated environments) matches the real car or environment.
There are examples of people playing with genetic algorithms to evolve a brain to control a self driving car although I seem to remember you saying the subject matter wasn't of any interest to you? I have observed over the years on this forum that there isn't any interest in advancing the construction and programming of a "real" working robot, self learning or otherwise.
@robotbuilder Correct, I have little interest in robotics but my curiosity was piqued by the contest and the lack of success especially when compared to Tesla etc. I wondered why Tesla was able to get 'it' to work and a bunch of bright college kids couldn't. Is this a problem of money or a skilled team vs one kid still learning?
My understanding of what Randell and Dennis are doing is that it is a new approach based on some math formulas as opposed to the vision (8 cameras) system + GPS that Tesla uses. I have never heard what kind of computer power a Tesla has, but it must be on the order of 10 or more equivalent Pi's or ESP32's and Randell and Dennis want to get something that will run on a single MCU/MPU like esp32.
Am I hopelessly confused?
First computer 1959. Retired from my own computer company 2004.
Hardware - Expert in 1401, and 360, fairly knowledge in PC plus numerous MPU's and MCU's
Major Languages - Machine language, 360 Macro Assembler, Intel Assembler, PL/I and PL1, Pascal, Basic, C plus numerous job control and scripting languages.
My personal scorecard is now 1 PC hardware fix (circa 1982), 1 open source fix (at age 82), and 2 zero day bugs in a major OS.
@robotbuilder Read the comments of that video, this is a sales pitch and he appears to be a scammer since the people who asked him where the free download was were ignored.
First computer 1959. Retired from my own computer company 2004.
Hardware - Expert in 1401, and 360, fairly knowledge in PC plus numerous MPU's and MCU's
Major Languages - Machine language, 360 Macro Assembler, Intel Assembler, PL/I and PL1, Pascal, Basic, C plus numerous job control and scripting languages.
My personal scorecard is now 1 PC hardware fix (circa 1982), 1 open source fix (at age 82), and 2 zero day bugs in a major OS.
@robotbuilder LOL, check the number of views and the comment count on all 3 of his videos. No comments on #2, and under 1k views a year on for number 3. That lack of interest tells us something I think.
First computer 1959. Retired from my own computer company 2004.
Hardware - Expert in 1401, and 360, fairly knowledge in PC plus numerous MPU's and MCU's
Major Languages - Machine language, 360 Macro Assembler, Intel Assembler, PL/I and PL1, Pascal, Basic, C plus numerous job control and scripting languages.
My personal scorecard is now 1 PC hardware fix (circa 1982), 1 open source fix (at age 82), and 2 zero day bugs in a major OS.
I wondered why Tesla was able to get 'it' to work and a bunch of bright college kids couldn't. Is this a problem of money or a skilled team vs one kid still learning?
Perhaps one is research and the other is a pragmatic practical attempt at getting something that actually works which means not trying to evolve a complete stand alone neural net brain.
My understanding of what Randell and Dennis are doing is that it is a new approach based on some math formulas as opposed to the vision (8 cameras) system + GPS that Tesla uses.
It is not a new approach. Evolutionary algorithms have been around for a long time as have neural networks. The breakthrough with neural networks is the back propagation algorithm and super fast, large memory computers to run them in real time.
The math formula is just a computation that the researcher believes will best measure the degree of success for each robot, a fitness value. This is used to decide the best, say top 10, to use to breed a new set of bots. I just sorted 1000 trial sim bots by how far they travelled and selected the top 10 that travelled the furthest without hitting anything. In nature the fitness value is reproductive success.
Am I hopelessly confused?
Not sure why? A real car navigating a real world is far more complex than those simple examples of cars learning not to hit walls. As I wrote above I don't see using real cars to evolve any kind of self driving car as being practical as a complete solution. Look at the attrition of billions of animals over millions of years required to evolve a biological brain. As far as I know current self driving cars are hard coded even if they use trained RNNs to recognize signs and other things captured by a camera as part of the program that controls the car.
Read the comments of that video, this is a sales pitch and he appears to be a scammer since the people who asked him where the free download was were ignored.
Ok. If that one was a scam I don't know. I am not interested enough to search for bone fide current research in the use of genetic algorithms but I would be surprised if they are all scams. My interest in the use of genetic algorithms was renewed for a short while with randell and inq projects and wrote a lot of code to test and analyze the results using sim bots.
There are practical uses of genetic algorithms.
https://en.wikipedia.org/wiki/Genetic_algorithm
@robotbuilder I think it best we drop it here.
First computer 1959. Retired from my own computer company 2004.
Hardware - Expert in 1401, and 360, fairly knowledge in PC plus numerous MPU's and MCU's
Major Languages - Machine language, 360 Macro Assembler, Intel Assembler, PL/I and PL1, Pascal, Basic, C plus numerous job control and scripting languages.
My personal scorecard is now 1 PC hardware fix (circa 1982), 1 open source fix (at age 82), and 2 zero day bugs in a major OS.
I didn't want to hijack the thread which is about the progress being made in a particular experiment.
Thank You
I think the experiment in randell's thread is seeing if he can evolve a set of weights using genetic algorithms for a neural net that will produce a behavior (obstacle avoidance in this case) given a set of inputs (distance measuring sensors) and a set of outputs (two motors).
Well Said
To err is human.
To really foul up, use a computer.