LRTA logo Light Rail Transit Association
Light Rail for better public transport


Author: Michael Bell


One of the problems in running public transport is staff. They cost a lot and they are unreliable. It is true that for a large vehicle carrying several hundred people, the cost of the driver is a small fraction of the total cost, but few urban public transport services have that level of traffic. Mostly we want to run services on which the average load is in the lower dozens and at that level the cost of the driver is the dominant cost. It is also difficult to get drivers to behave as they ought to and except in times of moderately severe unemployment it can be difficult to get staff at all.

Solutions to the problem have been sought in wholly segregated automatic systems. However this means underground tunnels or overhead ways, which the traffic cannot pay for, or very securely fenced surface paths, which would also be very expensive and very little less disruptive than the roads which public transport is supposed to avoid the need for.

This drives us back to consider the idea of running automatic transport along ordinary streets.

When we talk about "Public transport" in Britain it almost always means buses. They are by far the most important means of public transport in this country, they serve miles of street which we cannot think of leaving without service, yet these streets are almost always too narrow to lay out exclusive public transport lanes, and even if it was possible, frequent level crossings would inevitable and trespass impossible to prevent.

Therefore the new form of transport must be a bus replacement. Although in theory it must be possible to build a machine which could drive a bus (if a machine made of neurons can do it, why not a machine made of transistors?), I think it would be too difficult, but I think a machine to drive a tram could be built now.


It is no new thing to build computers to handle such situations, and to place reliance on them. In 3-dimensional space, air traffic control systems try to avoid collisions and military missile systems try to produce collisions. In 2-dimensional space, the Automatic Radar Plotting Aid (ARPA) must by law be fitted to ships over 10 000 tonnes, it gives warning to the officer of the watch that another ship is on a collision course with his own ship, and in principle it could be linked to the automatic helmsman to give fully automatic collision avoidance in accordance with the International Rules for the Avoidance of Collision at Sea. In "Logical space" there are computerised chess playing programs which can play to grandmaster standard, and much smaller and cheaper chess-playing machines which can beat the average player. This would be an acceptable standard for bus-driver replacement! A tram is fixed to one path by its rails. It operates in one dimension, and it is surrounded by vehicles and pedestrians moving in 2 dimensions. The automated tram driving problem is well within what has been achieved in other fields.


When we want to develop an automatic device to carry out a human function, it is helpful to study how a man does it. The task is "LEARNing to drive" (Not "drivING") and it is shown schematically in figure 1

The scheme is based on a division of the mind into two parts, a quick reacting part on the left, which is not capable of taking thought, but can react in accordance with programs entered into it by the second part on the right, which can think, but is too slow to keep up with driving situations as they occur.

As we all remember, when a man begins to learn to drive, all situations are new to him, and he has to THINK what to do each time a situation comes up, so he is slow and unsure in dealing with situations. He thinks about his near misses and makes plans to do better next time. As his experience builds up he recognises situations that have come up before, and if what he did last time didn't work well, he will try something different this time, and repeat the process till he arrives at an action that works well. So he builds up a repertoire of standard situations to be recognised and ways of dealing with them, he becomes faster and more competent in coping. If this were not so, we could not account for the fact that performance improves with experience.

This model makes explicit a very important point, that a man does not react to driving situations by "thinking" about them (in the way that a chess player "thinks" about a chess problem) and then taking action in accordance with the outcome of his thinking: the time available in driving situations is too short for that: Rather, he reacts automatically in accordance with a program he has created with his thinking slower brain. If a man can drive automatically without "thinking" (as we all do), why can't a machine? In this project the right hand side of the figure 1 represents the development team, who build the left hand side and modify it in the light of experience. The machine is the left hand side of figure 1, it is incapable of learning, but there will be a system (discussed later) which records accidents and near-misses for analysis and improvements to the program.


In developing the system I make the following assumptions:

  1. Satisfactory systems exist and are in service for railway-like functions like routing, signaling, stopping correctly at stops, door opening and closing, and I will not discuss these functions further.
  2. A tram can never be placed in a situation such that it needs to speed up to avoid an accident. Rubber-tyred vehicles can be placed in such a situation when they are overtaking, or to be more general, they can be placed in a situation where they need to speed up to make a turn which will get them out of trouble. The situation which can be imagined, where another vehicle is approaching the tram's route at right angles, and the tram has to decide whether to speed up to pass in front of it, or slow down to allow the other vehicle to pass in front of it, is unrealistic: It implies such a wide road that there would certainly be a reservation for the tram rails. This means that a possible organisation of the automatic driver is to calculate the highest speed that various elements of its situation allow and go at the lowest of those speeds, on the basis that it can never result in danger to go slowly.
  3. The tram has right of way over all other traffic. In many situations this happens automatically because the tram is on the main road, but it may have to be reinforced in places by having the tram set traffic lights in its favour as it approaches, by setting up "Give way" signs to protect its route or more simply by rewriting the Highway Code to give trams right of way.

    To give public transport right of way over other traffic is now almost traditional thinking to save time, but its advantages for a automatic tram go far beyond time saving. It does away with the need to observe the other vehicle's signals, which would be technically difficult to do, and it would be very difficult to interpret them and act appropriately on them. To give the tram right of way cuts through all these difficulties: The other vehicle either is or is not trespassing on the tram's right of way, and if it is, the tram must warn it off by sounding its warning sound, or braking: Any signals the other vehicle might be making are irrelevant.
  4. It is not possible to design a driving policy that can always avoid a collision. The best that can be done is to create a policy that will minimise risks of collision.

    The belief seems to be held by many that it ought to be possible to create a driving policy which will never allow an accident in any circumstances. This belief is often summed up in the slogan "Always drive in such a way that you can avoid the other fool's mistakes". While this may be an excellent frame of mind to be in while driving, it is a quite impossible policy to take literally, as a simple example will show.

    A vehicle, A, is driving along a main road. A second vehicle, B, is waiting at a side-road. A has right of way, so it can drive on. But it is possible that B will pull out. What should A's driver (man or machine) do? If A slows down as a precaution, then A will never get past B, in fact slowing down like this would invite B's human driver to pull out. But in the practical case, A's driver uses his commonsense and drives on.

    We have to accept that mistakes can be made which make collisions inevitable and there is nothing that another driver, man or machine, can do to avoid them. The best that can be done is to program the automatic driver to accept similar risks to those that a human driver accepts.

    I will discuss only the problems of "traditional" tram driving, two tracks, one each way, in a general traffic street and assume that if these problems can be solved, then so can those of various kinds of reservations and priority lane.

    In all discussion that follows, where I propose a particular solution to a particular problem, I do so only to show that there is at least one solution to that problem, not that the solution I propose is the only or best solution. Likewise, where I suggest values of time, distance, etc, I do so only to give the reader an idea of order of magnitude, and do not claim that that exact value is necessary.


The equipment fitted to the tram will be:-


Figure 2 shows that main organisation of the robot tram driver. At the top is the main input of information, the radars, giro, distance measurer and clock. The information from the radar is put into fixed frame of reference after allowing for the tram's forward movement, turn and sway by using the information from the giro, distance measurer and clock. A complete 3-dimensional representation of the scene is built up and 3 versions of it are held, the situation this scan, last scan and 2 scans ago.

The radar returns are analysed as presence or absence of an object in 10x10x10cm cells, and classfied as follows:-

The road surface. Any object rising above it is an object of some kind. Slope is observed to calculate braking effort.

The rails, embedded in it. The tram's path is defined by the rails plus the tram's overhang. The robot driver must know the extent and position of its path if it is to calculate whether another object is going to trespass onto it.

The pavement and islands. These are marked out from the road surface by the kerb, and it may be necessary to confirm this boundary by putting identifying reflectors on it. Projections of vehicle movement will assume that the kerb will not be crossed, unless implausible accelerations would be required to avoid it. (Whenever I mention "acceleration" I include "deceleration")

Street furniture may be marked by identifying reflectors.

Trees and lampposts are defined as having widths of less than 0.5M at ground level and rising above the upper height recorded, say 4M. They may be marked by identifying reflectors.

Persons are defined as defined in figure 3, between 0.75m high (the smallest baby that can stand) and 2.2m high, and between 1/4 and 1/8 of their height wide.

Vehicles are all objects not fitting the above criteria. Prams and cyclists count as vehicles.

3 sucessive scans are held. It may well be possible to store and handle the information more economically or more effectively than this but I will discuss it in these terms because the logical principles can be brought out more clearly, and it is a strucure which could be seriously considered.

Next follows the picking-out programs which pick out those objects which have moved between scans. One observation of an object establishes its position, two and a time measurement between them establishes the speed of that object and three observations with the time intervals between them establishes its acceleration, by which I mean not only its acceleration (or deceleration) in line of travel, but also turning accelerations. This applies not only to whole objects, but also to hinged objects, such as doors and articulated vehicles. The future path of all objects can be predicted from these three observations. Persons are of special interest and they can be identified as all objects fitting the frame of figure 3. (Some objects, mostly street furniture also fit this frame, but they can be separately identified by attaching identifying reflectors to them. This test errs on the side of caution by including too much rather than too little) People have to lean forwards to accelerate strongly, the effect is well shown in photos of sprinters. This feature can be used to estimate the acceleration of persons without waiting for three scans.

Figure 4

The various situation programs are then identified and dealt with as laid down in the rules. The definitions are defined so that they overlap slightly. For example, part of the definition of situation 1, the follow my leader situation is that there is a threat object in front of the tram moving in the same direction as the tram at an angle of less than 20° to the tram's path. Situation 3, the threat object from the left situation, is defined as any object coming from the left projected to cross the tram's path at an angle of more than 15°. Any object which falls into both definitions will be considered under both sets of rules and the more restrictive output applied. The general warning/braking policy for handling all situations is as follows:-

  1. On projected courses, deceleration of less than 0.05g is equired to avoid collision - do nothing, the situation is not urgent and may resolve itself.

    (Whenever I use the word "deceleration of xg", I also mean "deceleration or turn of xg" where it applies to a rubber-tyred vehicle. For shortness I use the single word "deceleration" to cover both)
  2. On projected courses, deceleration of 0.2 - 0.05g by another vehicle is required to avoid collision - sound the warning horn, the tram has right of way and must insist on it. If deceleration by another vehicle cannot prevent collision (eg when the tram is catching up behind another vehicle, then brake by as much as needed to avoid collision by 2M
  3. On projected courses, deceleration of 0.2g or greater is required to avoid collision - Sound anger siren and brake as required to stop 1m before projected collsion point.

It is now fairly simple to write the situation rules. Many of them come directly from the Highway Code, for example situation 1, the follow my leader situation. There is a vast body of mathematical analysis of this one. The others are simple enough once you think about them.

Situation 1. Follow my leader.


Situation 3 A threat object from the left


Situation 7a A threat object from the left is crossing the path in front of the tram, but the threat object will not be able to clear the path because another object is coming the other way. (Another situation, 7b, exists, where there will be room for the threat object to stand between the tram's path and the path of the object coming the other way.)


Situation 16. An oncoming threat object swerves out from behind a slower vehicle or queue of vehicles. (Situation 9 is the situation after it has swerved out)


By now the reader will have seen the kind of thinking used and be able to work out the rest for himself.


One doubt that will trouble many at this point is that human beings have better senses than a machine. This is not true in some respects, with its radar a robot tram will have much better information about speeds and distances (human judgement of these is notoriously poor), but it is true in some respects. For example a human being can judge from the position of a pedestrian's head or his stance whether he has seen a danger. This sort of clue is useful to a human driver because a human cannot scan a complete scene in less than about 1.5 seconds: Clues such as this enable him to concentrate on the danger points. Thus, a human facility is used to offset a human shortcoming. However, a robot tram driver can scan the whole scene in 1/3 second and no matter what the internal state of mind of another driver may be, he has to make an externally observable movement to create a new situation and such movements can be detected very quickly by a robot tram's sensors.

Another doubt is about the reliability of the electronics. Various systems of self-checking, system duplication, even triplication, exist and are well-established.

Remember, this is a street vehicle. The passengers can pull the emergency handle to stop the tram at any time, and get out and walk away.


Road development will obviously be finished off by running robot trams in real service with a safety man/fault finder sitting at an emergency stop handle at the front. This will be a far more severe trial than any human driver gets. With human drivers, the examiner decides whether the candidate is safe on the basis of watching him perform for much less than an hour.

When no more faults are found we will have to let the tram run on its own. How much distance has to be run before we can be sure that the robot tram's accident rate is at least no worse than the buses it replaces?

In 1998 bus and coach accidents on "built-up roads" (ie, with a speed limit of 40mph or less) were as follows:-

1021 1909 48710 780Total accidents
3.137296 367Accidents per 100 000 000 kM
SOURCE: Table 38, Road accidents in Great Britain: 1998 (ISBN 0-11-552161-5)

The distance run on this category of road by all the 65 000 vehicles in this class was 3 200 000 000 kM. Probably about half of "buses and coaches" were service buses of the kind we are thinking of replacing, and their accident rate was probably lower than these overall figures indicate because they run at lower speeds and are driven by drivers who are more familiar with the routes than coach drivers.

Figures for "Damage only" accidents are not available, but in any case are irrelevant: Since no human injury has been caused, we need only balance the cost of the damage against the cost savings arising from using robot trams.

Reporting of fatalities is probably accurate, "Serious injuries" are not all "serious" in the everyday meaning of the word and we know that the less serious ones are underreported: In one hospital survey it was found that 21% of serious road accident injuries had not been reported and for pedal cyclists the figure was 59%

"Slight" injuries really are slight and underreporting is known to be very widespread. We cannot guess beforehand whether they would be more or less completely reported for robot trams. So we can only make safety comparisons on the basis of "Serious accidents", and we know that even they have their inaccuracies.

To show with any pretence of statistical validity that robot trams have at least as good an accident record as buses, they will have to be run for a distance in which a reasonable number of accidents would have been expected. Let us say, because it happens to fit round numbers, that we have to run the robot trams for 100 000 000 kM, a distance in which buses would be expected to have incurred 37 serious accidents and 3 deaths. (And even this is statistically very thin ice). A single tram might average 20kM/hr for 18 hrs/day = 130 000 kM/yr. 100 000 000 Km is a genuine astronomical distance, it is ¾ of the distance to the sun! To build up a distance of 100 000 000 Km would take 770 tram-years! We cannot run a "trial" of that length! There is no alternative to testing our robot tram and sending it out to face the traffic, just as we do with a human driver.


This will be an expensive project to develop, it is worthwhile to try to list, even roughly, the rewards.

The exact extent of the savings which would result are hard to work out. Drivers are paid for more hours than they actually work, and they incur other overheads too, they need buildings for rest and rostering, wages administration, supervision, etc. On the other hand, what costs would be incurred in maintaining the equipment?

And what imponderable gains might there be? Better timekeeping should be attractive to passengers. It is also important to set our minds free from the traditional patterns imposed by the needs of human drivers which now do not constrain us. Crewed vehicles tend to run between terminals because that fits the human need for a rest at the ends of the runs, to go to the toilet, to count the money and to do the paperwork. Routes laid out on this basis tend to be close near the city centre and leave large segments of the suburbs between the radial routes awkwardly served. These problems can be resolved by running radial routes out of the city centre, circumferentially through the suburbs and back by a different radial route to the city centre, thus giving good coverage of the territory of the city and good routes between suburbs, which are important traffic routes nowadays, but are poorly served by traditional layouts. There is no difficulty doing this with robot trams, they can be run continuously and the slack time which appears at the terminals in traditional layouts can be realised in another form: If a tram on a loop route is delayed so much that it cannot regain its timetable slot, that tram and all the others on that route can be slowed down by remote centralised control to drop back to the timetable slot behind them and so restore the timetable as seen by the public, most of whom would have no way of realising what had happened. It would be difficult to do this and many other things with crewed transport.

One of the reasons why buses are often so big is to make the best use of the expensive driver. Once he is gone, we could go to something much smaller like the Parry people mover and have a service every 2-3 minutes. That really would change things fundamentally.

Although it is nothing to do with the technical possibility of running a robot tram service, there might be fears about whether passengers would feel safe from disorderly behaviour. While recognising that the feeling of fear is often out of proportion to the real level of threat, there are things which can usefully be said.

Having a crewed vehicle can make no difference to those who are frightened to wait at the stops. There can be no substitute for good civil behaviour backed up by adequate policing of the street! But there are some technical fixes which might help in ensuring security inside the tram. Closed circuit television has been found to be effective in reducing crime and there is no reason why it should not be fitted inside trams. There could also be microphones so that the central controller could hear as well as see what was going on. It should also be possible to program the microphones to raise an alert if there is a sudden burst of shouting or screaming, or alternatively, a sudden hush, the cues which trigger us human beings to look round. And loudspeakers can be fitted so that the central controller can shout back. But the real solution has to be to get the police to accept responsibility for ensuring good order at tram stops and inside trams, like any other public space.

There are at least 5 000* trams in the world. If the development cost were spread over them all, the costs would not be very great. But if we are seriously thinking of converting bus routes to tram, the market widens enormously, there are about 30 000* urban and suburban buses in Britain alone.

The possibilities for goods transport should not be overlooked. In the 70s British Railways developed the idea of building a fleet of driverless single self-propelled "autowagons" for carrying containers. They would be loaded at automatic loading points (several dozen in a large city, an important part of the proposal) and find their own way over the network to their destinations where they would be unloaded. The proposal was dropped, but not before British Railways had satisfied themselves that it was possible and in particular that automatic loaders were possible. Plainly goods-carrying trams on a city tram network could provide an even better service and spur tracks could go into the customer's premises. For the trunk journey the containers would be transferred onto mainline container trains.

About "Platform Pages"

This page gives an individual LRTA Member or other person the opportunity to present their ideas or plans for transport policies or proposals for their town, city or nation.

The views stated in "Platform Pages" are entirely those of the authors concerned - publication on this website does not imply that the LRTA endorses or supports the views expressed. Where "Platform Pages" challenge official plans or policies, the LRTA offers the government or other official body concerned the opportunity to present its views and comments on this page as well.

Webmaster's comment:

You can send your views and comments on this "Platform Page" to - they will be noted by the LRTA webmaster and forwarded to the author for reply

"Platform Paper" No. 3: top