Category Archives: Artificial Intelligence

Making driverless cars change lanes more like human drivers do

Please share this article:

In the field of self-driving cars, algorithms for controlling lane changes are an important topic of study. But most existing lane-change algorithms have one of two drawbacks: Either they rely on detailed statistical models of the driving environment, which are difficult to assemble and too complex to analyze on the fly; or they’re so simple that they can lead to impractically conservative decisions, such as never changing lanes at all.

At the International Conference on Robotics and Automation tomorrow, researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) will present a new lane-change algorithm that splits the difference. It allows for more aggressive lane changes than the simple models do but relies only on immediate information about other vehicles’ directions and velocities to make decisions.

“The motivation is, ‘What can we do with as little information as possible?’” says Alyssa Pierson, a postdoc at CSAIL and first author on the new paper. “How can we have an autonomous vehicle behave as a human driver might behave? What is the minimum amount of information the car needs to elicit that human-like behavior?”

Pierson is joined on the paper by Daniela Rus, the Viterbi Professor of Electrical Engineering and Computer Science; Sertac Karaman, associate professor of aeronautics and astronautics; and Wilko Schwarting, a graduate student in electrical engineering and computer science.

“The optimization solution will ensure navigation with lane changes that can model an entire range of driving styles, from conservative to aggressive, with safety guarantees,” says Rus, who is the director of CSAIL.

One standard way for autonomous vehicles to avoid collisions is to calculate buffer zones around the other vehicles in the environment. The buffer zones describe not only the vehicles’ current positions but their likely future positions within some time frame. Planning lane changes then becomes a matter of simply staying out of other vehicles’ buffer zones.

For any given method of computing buffer zones, algorithm designers must prove that it guarantees collision avoidance, within the context of the mathematical model used to describe traffic patterns. That proof can be complex, so the optimal buffer zones are usually computed in advance. During operation, the autonomous vehicle then calls up the precomputed buffer zones that correspond to its situation.

The problem is that if traffic is fast enough and dense enough, precomputed buffer zones may be too restrictive. An autonomous vehicle will fail to change lanes at all, whereas a human driver would cheerfully zip around the roadway.

With the MIT researchers’ system, if the default buffer zones are leading to performance that’s far worse than a human driver’s, the system will compute new buffer zones on the fly — complete with proof of collision avoidance.

That approach depends on a mathematically efficient method of describing buffer zones, so that the collision-avoidance proof can be executed quickly. And that’s what the MIT researchers developed.

They begin with a so-called Gaussian distribution — the familiar bell-curve probability distribution. That distribution represents the current position of the car, factoring in both its length and the uncertainty of its location estimation.

Then, based on estimates of the car’s direction and velocity, the researchers’ system constructs a so-called logistic function. Multiplying the logistic function by the Gaussian distribution skews the distribution in the direction of the car’s movement, with higher speeds increasing the skew.

The skewed distribution defines the vehicle’s new buffer zone. But its mathematical description is so simple — using only a few equation variables — that the system can evaluate it on the fly.

The researchers tested their algorithm in a simulation including up to 16 autonomous cars driving in an environment with several hundred other vehicles.

“The autonomous vehicles were not in direct communication but ran the proposed algorithm in parallel without conflict or collisions,” explains Pierson. “Each car used a different risk threshold that produced a different driving style, allowing us to create conservative and aggressive drivers. Using the static, precomputed buffer zones would only allow for conservative driving, whereas our dynamic algorithm allows for a broader range of driving styles.”

This project was supported, in part, by the Toyota Research Institute and the Office of Naval Research.

Please share this article:

Source: http://news.mit.edu/2018/driverless-cars-change-lanes-like-human-drivers-0523

Artificial Intelligence in Formula 1 Strategy – Part 2/2

Please share this article:

In Artificial Intelligence in Formula 1 Strategy – Part 1 I discussed the motivation for using advanced artificial intelligence in Formula 1 strategy.

I looked at the scenarios of pit-stop timing, tyre choice and team orders to give a few examples as to what AI in F1 could focus on.

In this article I want to look at a few key areas Formula 1 needs to focus on to make the transition to more and better artificial intelligence applications for strategy:

  • Data
  • More Data
  • Even More Data
  • Machine Learning Models
  • Huge Investment in Compute Power

Data

The teams are currently collecting a plethora of data from each test, qualifying and race. This “big data” will just grow and grow and grow. And the more teams learn about the data requirements for AI the more data they will undoubtably collect.

A few examples of data the teams have are:

  • Telemetry data from onboard sensors measuring anything from speeds, vibrations/frequencies, temperatures, pressures, etc.
  • Driver input to steering, acceleration, braking and of course the less numeric, but still vital, verbal communications.
  • Track data with times captured every 200m (give or take) + lap times, top speeds, pit stop times, track temperatures, wind speeds, etc.

A sample of Ferrari’s driver comparison telemetry data visualised.

I won’t go into more detail here as it is the given and obvious basis the sport is already operating on.

More Data

Of course, no matter how much data you can collect in the real world, there is always the opportunity to create useful data.

Not every race progress and outcome thinkable can be measured – given the limited amount of races per year.

However teams can create race simulations with realistic parameters and random influences to simulate millions of races. This simulation data can be used as “input” into machine/deep learning systems.

For example, imagine a “game like” simulation of every race on the calendar with many potential technical variations, weather variations, temperature variations, random crash and safety car insertions, etc. You suddenly have the chance to have enough data for an algorithm to learn what outcome was best in which scenario without ever having seen that actual race happen in real life.

 It cannot be too hard to get data like this from pure simulations. This is actual data from the Monaco race in 2011.

It cannot be too hard to get data like this from pure simulations. This is actual data from the Monaco race in 2011.

The impact of fictitious strategy decisions can then be observed and understood in quantitative ways with prediction abilities that give the team, sometimes maybe unexpected, strategy suggestions for any given race situation.

Compared to real world data this of course has the opportunity to become a much larger set of information.

I have a strong suspicion AI based systems will do an incredible job with this data given they have the right kind of machine learning models paired with them.

Even More Data

This is where the teams need to be innovative. As humans we can hear something is wrong with a competitor’s car when it drives by, we can judge how bad flat spots are visually in the TV slow-motion footage, we can judge tyre wear visually, we can “somewhat” hear things computers will not immediately be able to comprehend.

 Of course you won't expect to see images as clear as this, but you will see how bad a flat spot is ... paid that image recognition with other data ... e viola.

Of course you won’t expect to see images as clear as this, but you will see how bad a flat spot is … paid that image recognition with other data … e viola.

But, what stops Formula 1 teams from shooting super-high resolution, high speed, images or videos of cars driving by? Pairing that data with their own to judge vibration levels caused by tyre flat spots (usually created when a driver locks the breaks) should not be too hard either. Installing high fidelity microphones to record sounds in so much more detail than the human ear and brain can handle and using that data in correlation with their data to predict changes to a competitor’s car performance. Natural language input from other team’s radio communications. Visually measuring brake performance via relative deceleration comparisons and thermal vision. The list is endless, so many things could be done beyond what is already happening to collect more valuable and actionable data.

This is where F1 teams can gain an edge … by being more creative and clever than others.

Machine Learning

I won’t elaborate too much here. It is a given that a sufficient quantity and quality of input data is required to then feed into a new set of machine learning systems to start gaining insights and reap the benefits of AI.

F! teams will likely need to think of AI as a new department at least, with it’s own R&D, it’s own facilities and it’s own world-class staff.

An additional upside is that F1 teams can independently become leaders in artificial intelligence in general. Technology Groups like McLaren could start rivalling companies like IBM with a similar offering to Watson with the emphasis on strategic decision making in business.

Compute Power

Naturally all of this will require several orders of magnitude more storage space, computational requirements, communications speeds, etc.

But all of these problems can fairly easily be overcome these days. The cloud has vast storage and compute power to offer. And I bet it is not too much of an issue to set up a bespoke datacenter for these needs as well. If you can afford to run a wind-tunnel 24/7 you can afford what it takes to do AI the right way.

The biggest burden here is cost – how much is 1s per lap worth vs how much would it cost to get that advantage with AI is what it all boils down to.


Summary

It will happen, it just simply has to happen. In a game of diminishing returns on the hardware and aerodynamics engineering side of things, other avenues need to be explored to gain an advantage.

Artificial Intelligence with Machine & Deep Learning in Formula 1 Strategy will provide unknown and unexpected insights as well as highlight new areas to be explored for maximum gain.

Artificial Intelligence is one of the most glaringly obvious places where Formula 1 teams can make substantial competitive gains by:

  • Reducing or even eliminating human error in some areas.
  • Making better strategy decisions and predictions when little data may be available or processing is just not fast or conclusive enough.
  • Exploring completely new areas where AI can offer advantages.
  • Assuring drivers that the artificial intelligence is unbiased and will always work towards the maximum gain of first, and foremost, the team … and then the drivers.

This is a team sport and the money is awarded at team level after all … so the teams will have to move fast and significantly invest in AI before someone else does.

Read Part One

Please share this article:

Source: https://www.artificial-intelligence.blog/news/artificial-intelligence-in-formula-1-strategy-part-2-of-2

Artificial Intelligence in Formula 1 Strategy – Part 1/2

Please share this article:

Formula 1 (F1) often calls itself “The Pinnacle of Motorsport”. Hundreds of millions of Dollars are poured into Formula 1 World Championship teams every year – each – not in total. With billions of Dollars spent and even more earnt every year (TV licenses, prize money, sponsorship deals, track tickets, merchandise, marketing of car manufacturer brands, etc.) it is no surprise that F1 teams shy no effort and cost to improve lap times by mere 1/100s or even 1/1000s of seconds.

Everything, absolutely everything, is scrutinised and optimised to levels other sports don’t typically see.

This is high tech in it’s most extreme sporting guise. Space age technology cannot be adopted fast enough and innovation is at it’s best.

Humans are the weak point in a system of efficiency and precision.

With all this focus on detail, efficiency and performance there are some very obvious weak points … humans.




2-f1-ai-strategy.jpg


3-formula-1-track-data-for-machine-learning.jpg


4-formula-1-ai-data-visualisation.jpg

This season is shaping up to be one of the best in recent history with the two leading teams, Ferrari (my personal favourite) and Mercedes, fighting for top honours early on.

The first few races of 2017 highlighted the key F1 strategy areas where humans can fail, or at least perform to lesser standards than maybe an AI or humans supported by artificial intelligence would.

Drivers make subjective observations about tyre-wear, teams respond with strategies and strategy adjustments, mechanics perform sub 2-second pit-stops with tyre changes, crashes and safety-car phases throw in additional uncertainties (even the type of safety car phase, virtual or real, make a difference) and variables the human mind simply cannot compute fast enough to deal with optimally all the time.

 F1 teams have an incredible amount of very precise track data and they know at any time precisely where their cars are (GPS) and what state they are in (telemetry sensors).

F1 teams have an incredible amount of very precise track data and they know at any time precisely where their cars are (GPS) and what state they are in (telemetry sensors).

The strategy teams at the race track and at the team’s HQ are constantly trying to predict the next best optimal move to improve their drivers’ positions.

A lot of computation power, bandwidth and no doubt very sophisticated software is involved, all geared towards forecasting/predicting the final outcome of the race.

Despite all of this attention to detail and high-tech paired with experience and skill … they get it “wrong” quite often.

I wonder how, and at times can’t stop to think what it would be like if they applied artificial intelligence and machine learning in real time with “deadly” accuracy.

Of course the current top teams are doing a lot with sophisticated/advanced AI already, or are they? Surely they must be, but it is as many other things in this sport top secret.

It is in the nature of the sport that team members switch teams fairly regularly so I can’t imagine if a team was gaining significant benefits from A.I. it would stay a secret for too long.

This is right now though an early adopter’s game … who comes first to make it work will make untold millions more than they already do … maybe a chance for one of the smaller teams who cannot spend such extreme manufacturing budgets to catch up by dominating the strategy game?

What would some of the first things to focus on be? The “low hanging fruit”, if you will.

Pit-Stop Timing

First of all teams could use AI, more specifically machine learning/deep learning to get better predictions of when best to stop the car to change tyres. These better predictions could happen from very early on in a race and become more and more refined with increasing confidence as the laps count down. Such a model could take a lot more parameters into account than current systems and human strategists already do, but it would have to be possible to introduce human judgement for on-the-fly adjustments of the model – like for example when the radio communications of another team indicate a varied situation from the nominal model, e.g. complaints about vibrations due to a flat spot.

To be clear, what I am suggesting here would require truly incredible amounts of race simulation data with far higher level of information detail and density – and of course far higher real-time compute capacity (which is readily available via the cloud).

Tyre Choice

With the choice of when to best box the car (yep, they call it “box” … has German origins) the decision what tyres to put on has to be made. Strategies and best tyre choice might very quickly vary from moment to moment, especially when the bulk of other cars are stopping on the same or similar laps, but also as another of many examples, if the weathers is changeable. Weather is an especially interesting aspect here, as AI could be used in many more interesting ways to take local short and medium term weather data into account.

Predicting the weather (and temperatures of air and track) is too much, you might say. If F1 teams could guarantee themselves a 1s per lap advantage by having the tyre choice right I can easily see them spending vast amounts of money to do that.  There is of course, as with everything in Formula One, a level at which returns are diminishing, no matter what you spend … so, that has to be kept in mind.

Team Orders

Often enough we see much faster drivers being stuck behind their team mate. They may be faster due to lower tyre-wear, technical issues of the driver ahead or a host of other factors. Often the teams decide to issue team orders, sometimes they don’t, but in all cases it is still up to the driver in front to let their colleague through in a reasonable timeframe. Drivers being racers and wanting to be die-hard competitive winners don’t always obey these orders because they, well, just don’t want to.

Often that harms the teams in terms of constructor’s points scored and usually both team members are worse off. There is a strong human/psychology factor here. Nobody wants to be second, in anything, ever, full stop. And nobody wants to be told by (humans) the team that they are not the favourite child of the moment.

A few years ago team orders were still banned, but teams would use them anyway … and people get upset.

If the drivers however knew that an artificial intelligence, rather than human decision making (which can always be challenged on the basis of past bad calls) was at the root of this “recommendation” they might be more willing to comply, especially if they trust the system to benefit them in future in the same unbiased way.

What is needed to make this shift in F1 strategy happen?

  1. Data
  2. More Data
  3. Even More Data
  4. Machine Learning Models
  5. Huge Investment in Compute Power

Read Part 2



Please share this article:

Source: https://www.artificial-intelligence.blog/news/artificial-intelligence-in-formula-1-strategy-part-1-of-2

Kurzweil Claims that the Singularity Will Happen by 2045

Please share this article:

2029 is the consistent date I have predicted for when an AI will pass a valid Turing test and therefore achieve human levels of intelligence. I have set the date 2045 for the ‘Singularity’ which is when we will multiply our effective intelligence a billion fold by merging with the intelligence we have created.

— Ray Kurzweil

The singularity is that point in time when all the advances in technology, particularly in artificial intelligence (AI), will lead to machines that are smarter than human beings. Kurzweil’s timetable for the singularity is consistent with other predictions,– notably those of Softbank CEO Masayoshi Son, who predicts that the dawn of super-intelligent machines will happen by 2047. But for Kurzweil, the process towards this singularity has already begun.

In the Facebook Live Studio with Ray & Amy Kurzweil

We’re live from #SXSW with cartoonist and author Amy Kurzweil Comix, and author, futurist, and inventor Ray Kurzweil, after their Keynote for a conversation with our host Shira Lazar.

Posted by SXSW on Monday, March 13, 2017

By 2029, computers will have human-level intelligence,

— Ray Kurzweil

To those who view this cybernetic society as more fantasy than future, Kurzweil pointing out that there are people with computers in their brains today — Parkinson’s patients. That’s how cybernetics is just getting its foot in the door, Kurzweil said. And, because it’s the nature of technology to improve, Kurzweil  predicts that during the 2030s some technology will be invented that can go inside your brain and help your memory.

Read More



Please share this article:

Source: https://www.artificial-intelligence.blog/news/kurzweil-claims-that-the-singularity-will-happen-by-2045

New ARM Chip Architecture Promises Big Boost to Artificial Intelligence

Please share this article:

Devices are requiring more and more computing power, but AI software is expected to push those demands even further. To address this shift, ARM is launching a major update to its chip architecture with what it calls DynamIQ.

Artificial intelligence is at the heart of everything digital around us, from smartphones and buildings to self-driving cars.

— ARM

ARM’s new central processing unit (or CPU) architecture will cluster multiple different processing cores together with each one tailored for right software, including a dedicated processor for handling AI algorithms. Chipmakers will be able to develop CPUs with up to eight cores. ARM will also be releasing software libraries to better run the most popular AI techniques on its processors.

ARM claims this approach will boost AI performance 50 times compared with its current chips over the next three to five years.

As performance demands increase, so does the need for efficiency. Boasting a powerful combination of advanced compute capabilities, flexible design options and smart power-saving features, DynamIQ accelerates innovation of the future.

— ARM

Learn More

Please share this article:

Source: https://www.artificial-intelligence.blog/news/new-arm-chip-architecture-promises-big-boost-to-artificial-intelligence