In studies on cognitive cars, it’s found that there are a number of maneuvers that an automatically-driven car is unable to do such as: a) Drive on an unknown road; b) Self parking in structured parks or on city streets; c) Adapt to unexpected road change or construction site; d) Cross roundabouts where there is no signal; e) Recognize police instructions where normal traffic signals are overtaken; f) Pass a toll station; g) Fill a gas tank; h) Give way to priority vehicles; i) Implement safe three-point turn on street; and j) Make safe emergent stops on highways.
What are your observations or suggestions?
As per my observation if you incorporate Artificial Intelligence with cognitive car, It is possible to solve few problems pointed out by you.
Hi Mr. Wang,
As per your question, all the listed items can be done by and autonomous car. The trade off is how much safety is the car's intelligence is ready to sacrifice and how much cost is it ready to add in order to achieve these goals. For example the military has many unmanned tanks in operations which can navigate in unknown terrains. But then they are also prepared for the risk in venturing into unknown land. Some of these even has sensors to check land mass/type before putting its wheel on such places. Presently these vehicles are already capable of climbing hills and stealth attacking their enemies.
Of course some of the infrastructure needs to be modified to accommodate the autonomous car, Like bright yellow strips / magnetic strip on road for maintaining lane discipline, modified fuel stations to automatically fill autonomous cars (Something similar to mid air refueling by military aircraft ). Gesture recognition for understanding police instructions, Using RFID to pay for toll and gas (i,e RFID mapped to a credit card), Sound recognition for understanding priority vehicles. Special road sign and road sign recognition software can make pit stop on highway possible.
There are many more solutions for each of the issues mentioned in your question, but yes autonomous cars have a long way to go before it becomes practical and ready for mass consumption.
As for a science pet project or POC, of course all the mentioned issue can be solved with present technologies.
Thanks
I believe one of the most important aspect will also be efficient path planning i.e. tracking the traffic and planning the shortest path accordingly. However, as Mr. Bishwajit said, it can be resolved using the artificial intelligence.
I agree to Mr. Pal that the difficulties shown by Mr. Wang in the course of the technical development will come also to economically satisfactory solutions. The problem of every automation nevertheless continues: an automatic can react only to such situations which the developer has as possible scenarios assumed. With it every automatic also shows an error probability which can be, however, if necessary better than those of the driver. Today all automatics (all assistance systems available today) are so planned that they must be activated by the driver or can be deactivated. If one assumes the error probability of the driver for highly skilled activities usually with 10 E-3 and asumes in favor of the automatic that this one ten power is better, the overall error probability remains on the level of the driver because of this switch function. Therefore safety-technical seen no advantage by automatically moving vehicles is achieved in contrast to today in the press over and over again published statements. If one switches, however, the automatic function and the driver really in parallel, this overall error probability can be reduced significantly to 10 E-7. This is reached if the automatic system is switched on continuously, the driver, however, as before steers the vehicle in his own responsibility. If he bumps on borders given by the automatic function, his attention is drawn on it by suitable tips in the control elements (counter moment in the steering wheel or accelerator). Then from ergonomic point of view should be also signaled to him, why this intervention occurs. This happens best of all in the Head-Up-Display. After our experience the intervention of the assistance system by most drivers is not accepted without such a display of the Why. Of course the driver must be able to superpose the recommendation of the automatic system.
The problem of how to design a self-driving vehicle that can do traffic circles is difficult, and deserves a separate section. This is because first the vehicle must recognize that it is going into a traffic circle, then it must choose the best lane according to the unique right-of-way conditions that circles have, and then it must signal it's choice to leave the circle at the correct time. While each of these is relatively simple, they all fly in the face of standard lane logic, and the computer must make the decision to bypass it.
Mr. Salustri, I would like to defend the intelligence here a little bit. While I agree to your examples, which do not need intelligence but instincts, I also know the flexibility that will be provided by the Intelligent Brains of the control system. While the instincts can be robust, the intelligence of the brain can be used to a much larger extent. It purely depends on what exactly your requirements are.
Instinct is a biological imperative. Those ants with instincts can do nothing else. Intelligence is the ability to choose, to ignore or try to control the biological instincts.
It seems that the expectation is a little bit over optimistic for a practically machine-driven car using current AI technologies. Consider why we didn’t see any automatic train in real world, except in some closed airport terminals, where the driving environment would even be much simpler and controllable than those of cars on the road.
The extreme real-time complexity and uncertainty for a car driver, either a man or a machine, might not be under-estimated.
The answer may be to change the problem.
For example, in the 1950's britain, people envisaged that a dishwasher would be a robot doing what a maid did, I.e standing at the sink washing the dishes by hand.
What we ended up with was a box that you put the dishes into that when washes them. This was cheaper and didn't require the solving of any intractable AI problems.
The solution to autonomous cars may follow the same pragmatic
(Continued)...path, whereby there the majority of roads are modified to be constrained autonomous car friendly roads, and only back roads require manual driving.
This is analogous to manually washing a few delicate items that are not dishwasher safe, whilst most items are washed in the machine.
Dear all!
You answer to the question:What are the technical challenges to unmanned cars?
It is similar to the old discussion: horse or car.
What are the social consequences of unmanned car ?
The question is not how to drive the car but what the goal of a passenger.
All the features you mentioned can be incorporated in an unmanned car using the modern sensor technology and AI techniques.
The technologies in use exist, e.g. sonar, radar, GPS, the challenge is integration, power, EMI/EMC protection, risk management - i.e. how does a car fail safe, does it default to passenger or move off of road. Also have to address cyber security and protect vehicles from being hacked.
@Nicholas: > The answer may be to change the problem.
How would you change the problem? Change the car to a train or a horse?
A driverless car is a critical AI challenge that is much more complicated than that of a dish washer as you suggested. Consider the situation that the autonomous car is approaching to a cross and it has only 10s to change the lane while the target lane is already full of cars. How does a driverless car solve this kind of real-time decision making problem?
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Artificial Intelligence