Skip to content

Recent Comments



    This is very much a windshield perspective and seems to ignore the reality of other users. You are promoting all the speed and throughput dreams. The capabilities of these autonomous car fleets is theoretical at this stage. None of this improves safety. In my line a work we say to error is human, but to really $%^*&^* you need a computer. I strongly believe that with autonomous cars we may (if properly designed) improve safety. More likely we will simply be trading certain types of human caused fatalities for a different set of computer generate fatalities. And with even less accountability.

    High reliability computer systems that depend on sensors are very complex and expensive. Of course you could always trade cost for less reliability, assuming there an acceptable level of fatalities. What would that number be? How about zero?

    Personally I don’t even want to be near an autonomous car going 150 mph. There is little margin for error and a very high consequence if an error occurs.

    There is no practical way to avoid a protracted mixture of human and autonomous cars. I don’t see the government getting away with a mandate before widespread adoption and the resulting disenfranchisement of those who can’t afford all the new technology.

    When you stand back and listen to what the autonomous car people are saying, it sounds an awful lot like they want a really good public transportation system. We could do a lot now with professional drivers and do it cheaper.

    So how do autonomous cars fit into an ecosystem with pedestrians and bicyclists (and deer, potholes, ice, construction, high crosswinds, police directing traffic contradicting control devices, etc). Not easy and needs to be thoroughly thought through. USDOT is only just making the first baby steps.



    The issue with HOT lanes is that we are creating haves and have-nots in the highway system. It used to be that everyone, no matter their income level, had to sit in the same traffic on the highway. Now that HOT lanes exist, those that make more money can be assured free-flowing highways for a small fee while everyone else sits in congestion. The danger with this is that those with money are also typically those with power in our society. One reason why transit is so overlooked and underfunded in the US is that, for the most part, those with the power to affect change don’t use it. The general-purpose highway system risks losing influence in the same way and, therefore, funding if those with power/money abandon it and just zip to work on the HOT lanes.



    HOT lanes are just like HOV lanes – a way to add capacity in a fashion palatable to environmentalists and air quality regulators. HOT lanes will become congested and useless slower than a general purpose lane, but eventually will clog up as capacity is reached.

    I find enforcement on HOT lanes to be questionable at times, as the I-10 express lanes can attest. The tolls are sometimes up to $20 end to end but are still packed, and part of the reason is that people are cheating the system, and there is no automated means of enforcement. Motor officer enforcement is very tricky since you have to pass by a single occupant vehicle and verify the correct occupancy is set while they are passing by at speed. Rarely are HOV lanes totally underutilized, and the tolling mechanism is not dynamic enough to charge extremely high rates for the lane when necessary. HOT lanes can subsidize transit providers, which may be a net positive over the equivalent HOV lane if transit service would not be funded along the corridor, but is a negative if that service becomes more unreliable due to SOV toll payer caused congestion (i..e. LA Metro Silver Line).



    For those of us reading this in the rest of world, outside of the United States, 40 mph is 64 km/h, 30 mph is 48 km/h, 25 mph id 40 km/h, and 20 mph is 32 km/h.



    Riverside Freeway (aka 91) has a stretch of paid HOT lanes with variable pricing – sometimes it’s $15.00, sometimes it’s $2.00

    great system and should be applied statewide



    applause applause



    Another issue with HOT lanes is politicians receiving pressure to reduce or eliminate tolls during certain periods of time.

    For example: WSDOT’s I-405 HOT lanes. The two-year trial period of variable 24/7 tolls was reduced to six months after local politicians, who originally approved the project with a two-year period, backtracked after pressure from constituents. Now, the HOT are free and open to all users except during weekday peak times. With HOV/HOT conversion and new HOT lanes slated to expand, this scenario will likely repeat itself now that people know how to pressure their elected leaders.



    How about the concept of HOT cities, and surcharging for use of those?



    One thing that can be said for HOT lanes is that they are an entry point for paid use of highways. Ultimately, it would make sense to toll ALL congested highways, even if it’s a nominal amount. In that case HOT lanes can continue to be the premium lane but with a price point that truly keeps HOVs and buses running smoothly. Any tolling scheme must aim to provide viable alternatives for those can’t (or don’t want) to pay the toll.


    Joe R.

    Being that I’m an electronics engineer, sooner or later I’m bound to get back to reality. Yes, this is a difficult but interesting problem. The nice part is as you said the system will evolve. The software will doubtless be updated regularly as we learn more, making it better as we go along. Strictly speaking it’ll never be a done deal. I would imagine even if we reach a point where we go from 35K traffic deaths down to single digits we’ll still try to aim for zero. And once that happens we’ll doubtless start analyzing close calls which don’t kill anyone just to prevent those.



    Aah, now you’re talking real-world!

    I find both parts (detection and handling) to be fascinating and exciting. The interesting part about the algorithm is that we expect it to learn from experience and improve – to be able to go beyond merely the situations it has previously encountered in test runs and handle somewhat similar situations with a modicum of intelligence. And we expect the learning to be collective across the system rather than restricted to individual agents.

    How to translate information from the sensors to a world model. How best to represent and manipulate this knowledge system? How to share it? How to digest and re-assimilate across agents, so that each one can operate offline when required? May be not as critical as hazard detection, but fascinating problems all the same. I am sure people are working on these and more. But they’re far from a done deal.


    Joe R.

    It’s a matter of breaking them down into algorithms. To offer some insight, let’s take what I do in similar situations while riding my bike. I’ll perceive a potential situation up ahead, then adjust my speed and course to avoid worst case scenarios. For example, suppose I see a child on the sidewalk. My first reaction as I get to within about 100 feet would be to start moving as far left as possible to give myself more time to react should that child suddenly decide to dart out. As I get closer I do some quick mental calculations, as in if that child were to dart out in front of me right now, could I stop in time or avoid them at my present speed? If not I reduce my speed. Something like this isn’t terribly hard to program.

    If you ask me, I think the primary problem here isn’t one of figuring out what to do once potential hazards are known. Rather, it’s detecting those hazards in the first place. Tesla screwed up big time in that area with the truck crash. Using multiple types of sensors and preprocessing the data will both increase the odds of early detection and reduce the chance of false positives. The latter is important as well. You don’t want an autonomous vehicle slamming on the brakes if a leaf blows in front of it.



    Slowing to 20mph doesn’t just magically solve problems.

    If you are driving along in a residential neighbourhood and see a football roll across the street, you have to be prepared for what comes next. You can’t just continue at 20mph. If you see an owner chasing after an off-leash pet on the side-walk 500′ up the road, you have to account for the inherent danger when you get close. See a deer browsing about on the side of the road – it might decide to run across. The car needs to perceive these situations and prepare for them – go beyond just slowing down to 20mph. If a collision could have been avoided, then the fact that it only happened at 20mph and did not cause loss of life is not a sufficient recourse. And then there is a host of road-conditions between urban roads and freeways – suburban, rural, back-roads, mountain terrain, etc.

    You keep missing the point – it is not a problem that can be solved ‘in the world of computers’ or ‘on a cloud’ – it is a problem that must be solved in the real world and on a massive scale.



    More ho-hum futuristic technology! Talk about it when it exists – when it is economically and technically feasible to deploy on a large scale. See how many new problems crop-up when you get beyond hand-waving and in to the nitty-gritty?

    > People smarter than both of us will most likely test for all the
    > things you’re worried about.
    Oh, I am sure they will! I worry about people lacking the smarts pushing those smart people in to doing stupid things in order to be able to deploy yesterday!


    Joe R.

    10 years is a really long time in the world of computers. And the pedestrian protection need not necessarily be airbags. There are lots of other possible solutions. Just slowing to 15 to 20 mph will solve a lot of problems. It would even if we could get human drivers to reliably do that but we can’t.


    Joe R.

    I’m thinking more from an acceptance/legal standpoint than a practical one. I’d personally be happy with autonomous cars which are as safe as a decent sober, nontired driver but the general public and lawyers might not be. Any crash will likely make headlines, as it did with the Tesla crash, even if the average crash rate was far lower than for human drivers. For those reasons, we probably need to be a few orders of magnitude safer. And I think that’s easily within the realm of possibility. Heck, I haven’t crashed on my bike in 2 decades and I’m far from being a computer. I just anticipate every possibility and give myself an out. Doubtless a computer would be much better at this than me.



    But even that technology (external airbags which operate reliably) doesn’t exist right now. What little does, cannot control where the victim will bounce off to after the collision. There is a big disconnect between your futuristic vision and current reality – when the disconnect is eliminated, then sure all those things are possibly. I’ll accept that these problems are solved when they are actually proven to be solved. A wave of your hand is not good enough!


    Joe R.

    Trucks will still be needed for last mile delivery from the rail depot to wherever they’re going but they need not be going long distances on Interstate highways where we would have the 150 mph speeds.

    You’re assuming vehicles need to come to a complete stop to avoid danger. Usually changing direction is sufficient. It’s not like a concrete barrier is going to suddenly appear across all lanes of a highway. Some lanes will still be clear if there’s debris. A human driver might not be able to react in time but an autonomous vehicle can. It can also most likely detect the obstacle before a human driver does due to multiple types of sensors.

    And don’t you think autonomous trucks would have some means of detecting if the load is coming lose before it actually does? Abnormal vibrations or small changes in power needed to maintain a given speed are all subtle clues a load is shifting. A human driver probably couldn’t detect those things. Heavier fines for unsecured loads could also mitigate the problem. Given that you no longer need a truck driver, loads would be secured at their point of origin and they should have to meet certain standards before the truck’s software will even allow it to proceed. You really can’t liken the situation to now where we have amateurs who don’t know what they’re doing putting loads on trucks. We may even design trucks so the loads can’t fall out if they become loose by totally enclosed them.

    300m+ to stop from 150 mph? What about using aero braking where you have drag-inducing flaps extend at high speeds and/or induce downforce? I’ve little doubt you can get the 150 to 0 braking distances down to 100 meters or so while keeping deceleration rates to a tolerable (for a few seconds anyway) 2g.

    The question is not whether it should be safe, but will it?

    People smarter than both of us will most likely test for all the things you’re worried about.



    They must also avoid crashes caused by animals or poor road conditions a few orders of magnitude better than human drivers.

    Is your reference driver a below average sober driver or an intoxicated/tired/angry/stressed driver who crashes far more than the ‘average’ driver per mile? Used judiciously, a driverless car equivalent in safety to the average sober/non tired driver could offer substantial safety benefits. If dealing with people not in cars who aren’t not following the law to the letter is possible, why would dealing with them in cars be so much more difficult?



    Rails are not point to point. Even when extensive rail networks, you’d still need freight trucks for the last mile problem. Not to mention – that kind of rail infrastructure would take decades to develop, even if the political will and money were at hand, which it isn’t. So lets just think in terms of the actual world we live in, what say?

    I’ve seen pickups drop garbage, and construction cones quite regularly on the free-way. Its not an out-there occurrence – it happens all the time. Sure it can be fixed, but these kind of upgrades needs both time and money.

    Also note that an 150mph there is less than half the time to react than there is at 70mph and space needed to stop increases quadratically. Doesn’t matter how instantaneously the autonomous vehicle brakes – it still needs 300m+ to stop from 150mph under ideal conditions, as does the vehicle behind it and the vehicle behind that one. The question is not whether it should be safe, but will it?


    Joe R.

    The crash of the Tesla was due to error on the part of the truck driver combined with a failure of the car to detect the truck. However, human error was the primary cause here. Had that truck been under autonomous control, it never would have changed lanes without first checking if it was safe to do so.

    I’ll guarantee that an autonomous car that *never ever* crashes in to another car, but occasionally kills pedestrians in cross-walks or cyclists following the law, will never be acceptable. Some of those secondary causes are deal-breakers and show-stoppers – partial solutions to those problems will not fly.

    Keeping speeds to 15 to 20 mph on urban streets, plus having some sort of external air bag which inflates while driving on those streets, should more or less eliminate killing cyclists or pedestrians. It’s not a hard problem to solve BUT you just can’t have cars going at high speeds once they’re off highways.



    It will be as short as ‘reasonably’ possible – corporations are plowing billions of dollars in to it. It may yet not be short enough for your liking – tough luck! Two things:

    1. Autonomous cars will be held to a higher standard than human drivers in terms of avoiding preventable crashes and dangers. Think of the safest human driver you know in terms of skill, then add unlimited patience and unlimited stamina!

    2. Public opinion will not be sold over by mere numbers or probability. Global reduction in danger is meaningless if instantaneous personal danger increases – i.e. few & spectacular failures weigh heavier on the public psyche than many small ones (think Tesla’s auto-pilot smashing you in to the side of a tractor trailer at 74mph in day-light).

    Push too hard to shorten the transition period and you reach a position where actual (or perceived, doesn’t really matter) risk of death by autonomous cars exceeds human driven – which in turn restricts the uptake and further lengthens the transition period.

    Finally, for all your belittling of ‘secondary causes’ – I’ll guarantee that an autonomous car that *never ever* crashes in to another car, but occasionally kills pedestrians in cross-walks or cyclists following the law, will never be acceptable. Some of those secondary causes are deal-breakers and show-stoppers – partial solutions to those problems will not fly.


    Joe R.

    Mechanical failure which can cause crashes is very rare these days. It would be even rarer once we have fleet autonomous vehicles which are held to higher maintenance standards. As for trucks, we shouldn’t even be using them for long distance freight. Put that freight on rails where it belongs and an issues caused by trucks, including heavy road damage, go away.

    Note that 150 mph under autonomous control should be far safer than 70 or 80 mph under human control even if things do go wrong. It takes an average human around a second to process danger ahead, a few tenths of a second to react to it, plus a few more tenths by the time the vehicle reacts due to the cumbersome controls needed to interface with a human. An autonomous vehicle processes the danger and gets the vehicle to react nearly instantaneously.

    Finally, higher speeds are a MAJOR selling point for autonomous vehicles. If we want to encourage people to give up control of their vehicles so we can ban human driving on public roads, we need a carrot as well as a stick. The stick would most likely be much higher insurance premiums for human driving. The carrot is much higher average travel speeds.



    if Paris can open her riverfront to people, why not Manhattan ? Close West Street to motor traffic every weekend.



    Yeah, right! What happens on the expressway when you’re going 150mph and the vehicle ahead of you has a mechanical failure or a cargo truck spills debris on the roadway? Would a 100 car pile-up be an acceptable outcome?


    Joe R.

    To a very large extent keeping maximum speeds in the range of 15 to 20 mph while operating around pedestrians or cyclists should reduce both crash probability and likelihood of injury/death. It’s a given that an autonomous vehicle will encounter unpredictability (pedestrians, cyclists, falling tree branches, black ice, animals to name a few). It must be able able to deal with these things. The first way is by detecting the possible danger in the first place. You can use some combination of radar, cameras, life sign detectors, infrared, microphones to see what’s out there. The further off you can reliably detect something without false positives, the fewer evasive maneuvers you’ll need to take. In the end this is probably the most challenging part of autonomous vehicle design. If all an autonomous vehicle encountered were other autonomous vehicles we probably could have had them on the roads using technology from a decade ago. Obviously that’s not the case. They must reliably avoid colliding with cyclists and pedestrians. They must also avoid crashes caused by animals or poor road conditions a few orders of magnitude better than human drivers.



    nearly all are caused by interacting with unpredictable human drivers.

    And human pedestrians and human cyclists.


    Joe R.

    The transition period needs to be as short as possible. If you look at the instances where autonomous cars can fail, nearly all are caused by interacting with unpredictable human drivers. There are of course the other things you mention, but those are a secondary cause of vehicle crashes.


    Joe R.

    By default we should mandate that all vehicles be electric within a decade at most. The technology is mature at this point. The only thing preventing it is lobbying by oil companies.


    Joe R.

    It also negates the need for traffic signals. Cyclists and pedestrians can just cross an intersection when they arrive at it, without even slowing down, confident that self-driving cars will yield to them.


    Joe R.

    Self-driving cars should result in MUCH higher expressway speed limits, probably on the order of 100 to 150 mph, once human-driven cars are out of the equation. Expressways are where the most time would be saved with higher speed limits anyway, not urban streets.

    Also, we should minimize any so-called transition period. Once self-driving cars prove themselves, human driving on public roads should be outlawed not long after. A mix of human and autonomous cars negates most of the advantages of autonomous cars.



    If you agree that there has to be a transition period and it will need technology capable of allowing autonomous and human drivers to mix, then why do you criticize Google for developing that technology? Do you think it is possible to get to your end-game scenario without the so-called transition? If not, then why is it ‘a problem’ that Google is developing it? Plan ahead all you want – but we ain’t gonna be able to build anything without a foundation.

    IMO, what you belittle as ‘transition technology’ is the gut of the system. It will always be relevant – even when there are no human drivers. There will always be cyclists, pedestrians, oil spills, fallen trees, animals running on to the roadway, network failures, rain, snow, black ice and so on and so forth. What the UT vision of the future lacks – but the real world contains oodles of is non-determinism. Planning for this non-determinism is the hard technical problem that must be (and is currently being) solved before wide-scale deployment of autonomous technology. Its a real technical problem – its not just lack of political will.


    D G Spencer Ludgate

    True, you may not start on the 101st floor; but you do start with a plan. In eight years we went from WWII/V-2 rocket technology to the moon. We achieved the lunar landing using a computer that ran at 1MHz with 2K memory. The technology being developed today is for the transition period when autonomous vehicles have to mix with human drivers. Today we have the computing technology to create the network and retrofit all vehicles – just not the political will to do it.



    I don’t see how lack of central control affects safety – even in your fantasy world. Efficiency – sure! But even to get to your fantasy world, there first need to be an automated vehicle capable of operating with minimal safety standard, even if in its own self interest. This is exactly what Google is doing or at least attempting.

    Until basic safety by individual agents is achieved, that video of UT is pure speculative bullshit. Might as well be watching Jetsons for all the good that vision of the future will do us. You don’t construct a high-rise by starting with 101st floor. You have to know where to start. You’re basically saying that time spent digging a foundation is waste because its going in the wrong direction!


    D G Spencer Ludgate

    My point is that we will not see safety and efficiency gains until all vehicles are autonomous and linked into a network. When I say networked, I am referring to communicating with other vehicles and traffic control devices. I uploaded a video from the University of Texas that uses networked computers at intersections.

    Let’s use this assumption. I hop into my autonomous vehicle. I enter my destination (e.g. my Hollywood Hills to Torrance commute). A pre-planned route is created. For the entire drive, it already has been determined what lanes, speed, spacing with others along the route, and even what intersections I will have to stop for pedestrians. Once that is calculated, I will be informed when I will enter the grid and when I will arrive at my destination. As I proceed to my destination. similar to a cellphone network, navigation gets passed from computer to computer.

    Think of it as a cross between cellphone service and air traffic control. When you travel by commercial aircraft, the pilot files a flight plan. The pilot then follows a pre-determined route. The local airport clears you for takeoff. Assuming no issues at the destination airport, once in the air you fly non-stop to the destination, and come to a landing without circling. En-route, the plane is handed off to regional air traffic controllers, and the destination airport handles the final leg and landing.

    So, the autonomous vehicle is the pilot and the networked traffic grid is air traffic control. And similar to air travel, “Congestion” will be the wait time to enter into the grid. You may hop into your car, set your destination, but have to wait ten minutes until you can enter the grid. Riders may be able to schedule a trip in advance to avoid waiting. (Using my commute example, I may set up a weekday reservation to arrive in Torrance at 6:55 am. I may have to be in my vehicle and ready to go at 6:22 am or I loose the reservation.)

    There needs to be some sort of central control, or you will end up with hundreds of thousands of autonomous vehicles operating in there own self interest; just like we have now with human drivers.



    No idea what you mean by ‘network level’. With that many agents and that much non-determinism, control has to be decentralized – which is exactly what Google is working towards. An automated vehicle may or may not talk to another vehicle or to the transport network – but it must be able to operate at a basic level of correctness and safety even in the absence of such side-band communication. Macro-level safety & efficiency may be improved by networking but micro-level safety is based on individual agents making limited assumptions about the behaviour of other agents.

    Incidentally, decentralized control is also a central feature of cellular networks, satellite networks, internet – you name it. There is no other way of achieving scalability and fault tolerance.



    fyi. The Wired article is from 2014. But it is still relevant today.


    Kenny Easwaran

    Within cities, actually obeying speed limits isn’t likely to make much difference to total travel time. On intercity rural expressways it would, but within cities travel time is dominated by signals and stop signs and even traffic, so that even a major drop in top speed from 35 mph to 25 mph wouldn’t make a huge change in travel times.



    Blocking a highway is intentionally holding people against their will. Traffic is not intentional. Someone could have a panic attack and drive through the people to escape.


    Jake Wegmann

    By “core,” I mean something beyond just the CBD proper, which yes indeed is full of parking lots as is the case in most other cities in the US. Really I mean inside the 610 loop. If you go to the Heights or Montrose you’ll see relatively middle-income households able to afford a close-in townhouse in a way you won’t see in many other prosperous US cities.



    Cool. But when I look at the core of Houston, I see parking lots. And parking garages. And more parking lots. And parking garages next to them.

    I’ll believe something is happening when the parking lots start being built on.



    1. Partner with ride-hailing services
    Ride-hailing services need to require wheelchair-accessible vehicles before this will work.

    2. Modernize ride reservation and fare payment systems
    Not gonna help.

    3. Provide real-time information

    Will increase costs, will not reduce usage.

    4. Right-size vehicles
    Make wheelchair-accessible taxis universal (like London, UK did in 1989) and this becomes much more *possible*.



    Sure, it would be great if self driving cars would drive the posted speed limit, follow all the rules, and make things safer. That is is what they are selling to help us all buy into this idea. But can you imagine the disappointment self driving car owners will have when it takes forever to get anywhere because their vehicle follows all the rules? Especially while there is a mixture of computer and human drivers. All the human drivers will be passing them in droves and the human passengers who are insufficiently passified will get resentful.

    One possible outcome is that the speed limits will change. Maybe other rules too. Remember it all about throughput in the end. That’s our deeply ingrained transportation culture. Does anyone think this culture is really going to change as quickly as they say these things will be deployed? Sure safety is being pushed now and it sounds fantastic. Once this is established I can see things changing in ways that other road users will not be happy with at all.

    Right now speed limits are set using the 85th percentile rule, based on the speed at which drivers (humans) go with some perceived level of safety. (I know this already disregards other road users.) So how would new speed limits be conceived if there are no human drivers? Will it be based on the capability of the self driving computer to be safe? We are told self driving cars can “think” and react much faster than the humans. I can image that capability being used to make cars go even faster because it is still safer than the human driver. This will be really bad for other road users who will be even more marginalized than before.

    Can you see the all powerful bike lobby flighting the self driver car interests with all there vast investments and hype?

    Beware of the law of unintended consequences as they loom large with self driving cars.



    “Yield to Pedestrians and Bicyclists at Intersections and Crosswalks;”

    Does this mean that “jay”walkers are fair game? Death Race 2000! And surely bicyclists are expected on the road outside of crosswalks.

    It would be easier to just remove the “at Intersections and Crosswalks” clause and simply yield to pedestrians and bicyclists anywhere on the roadway. That negates the need to differentiate between crosswalks and anywhere else on the road.



    Agreed, hence the qualifier “live and work in the city”. Lots of people commute across city borders in one direction or the other, but still there must be a good number within the city. Of course, the issue of hostile roads discourages many of those people even if their commute is only a couple of miles.


    Joe Biel

    Yes, like many post-industrial cities in the U.S., Reading’s affluent white population fled to the suburbs and exurbs. LOTS of the white commuters that we interviewed do cycle from the suburbs into the city.



    A major problem with cities of this size is that a substantial number of jobs are now located on the periphery in exurban shopping centers and office parks that are absolutely horrendous for walking and cycling. The video focuses on cycling in the urban center, but I’m curious how many commutes go outside this area (on a county level, this map shows that Berks County residents scatter pretty far).



    It’s really great to see a city focusing on people (low-income, minorities) who already bike, recognizing that their needs and safety matter. I’m curious if they’re facing any bikelash, particularly with measures like replacing city parking permits with bikeshare (an excellent move).



    Cities under 100,000 people tend to be small enough that you can bike between the farthest two points in the city in about 30 minutes. That should make cycling an eminently practical mode for people who live and work in the city. Too sad that it doesn’t happen more often.



    List of car-free places
    From Wikipedia, the free encyclopedia