|
Time to fly on autopilot?
|
01/04/2007
|
| |
Technically there are few bars to giving a vehicle complete control of cruise control, down to stop-start in urban environments. But it is not happening. Jeff Daniels looks at the reasons why.
Our industry faces an intriguing paradox. On the one hand, it is generally acknowledged that human failure on the part of the driver contributes to around 90% of all injury accidents. On the other, technology already exists to enable vehicles to be automatically controlled, at least in some situations, using sensor data free from human failings.
The paradox is that thus far the industry has set its face against the concept of a ‘vehicle autopilot’, despite the arguable safety advantages. It always argues that ultimate responsibility must remain with the driver, and the authority of existing systems is limited to the extent that, while fully automatic operation is possible in free-flowing motorway traffic, the crucial phase of braking to a full stop requires driver intervention. The motivating force behind this position is fear of product liability litigation.
Today we are on the verge of systems that will allow stop-start operation in urban traffic, and at that point manufacturers will either have to accept that some drivers will employ ‘autopilot mode’ by default, or devote some technical ingenuity to devising ways of preventing it. An example was provided by BMW, which some years ago demonstrated an assisted lane-keeping system that would only work if the driver was gripping the steering wheel.
No aviation style autopilot
This kind of thinking puts the philosophy ahead of the technology. Despite the common use of the term ‘autopilot’, familiar from aviation, the vehicle requirement is fundamentally different. The function of an aircraft autopilot is to hold a user-defined steady-state situation: direct pilot intervention is called for only in emergency, for example for collision avoidance, or where autopilot performance is limited by data quality, such as when landing. Most airliners complete 95% of their flights under autopilot control. In a road vehicle, an autopilot function must depend on two distinct abilities, the automatic tracking of a lane, and the maintenance of safe separation between the user and any other traffic in that lane.
A decade ago, lane-tracking systems were being designed around leader cables or passive studs in the road surface – expensive to install but simple and reliable. This was technology borrowed from existing automatically guided vehicles in factories (not least automotive body shops) and other ‘closed’ situations. It remains an option, especially for guided-bus systems, but for most purposes it has been superseded by CCTV camera-based systems and image processing.
This technology has advanced on two fronts. In the mid-1990s the sensor cameras were expensive and image processing depended on extremely clear and precisely applied road surface markings. Today, suitable cameras cost less than 100 euros and image processing is capable of deducing the optimum path along almost any road with a marked centre-line and some form of edge indication.
Modern systems, already offered in production cars, no longer have any problem calculating an ideal lane-keeping path on any main road. The remaining challenges arise at junctions and on country lanes. This is one of the limitations which continues to drive research into the alternative of extremely high precision GPS as the primary sensor.
The question of the moment concerns the form of the system’s output. If the lane-keeping system is purely ‘driver advisory’, the output takes the form of a warning, which may be a shaking of the steering wheel, a ‘nudging’ from one side of the seat cushion (the option chosen by PSA for its system), or the generation of a ‘rumble strip’ sound and feel in the steering wheel, the approach adopted by BMW in the lane departure warning system now offered in the new 5 Series, using a system developed by Siemens in conjunction with intelligent-vision specialist Mobileye NV. This system provides progressive warning as the demarcation line is approached, rather than acting only as it is crossed.
The significant move forward comes with a system output directly to the steering. This may be achieved with a ‘strap-on’ electric motor, or in the case of cars with electric power assisted steering (EPAS), with a direct input to the steering motor. In either case, in all current installations the maximum steering torque is limited to that needed for a lane-maintaining correction – up to 10Nm. This is small enough to be easily overridden by the driver, if for example he wishes to change lane. The system may in any case be inhibited by operation of the indicators, re-engaging once the vehicle is in a new lane.
The small lane-maintaining torque may not suffice when entering a relatively sharp bend, so some form of driver warning that the system has hit the limit of its authority is needed. This also takes care of cases where, for example, the lane markings confuse the system, perhaps where two lanes merge into one. To enable a vehicle to follow a fully plotted ‘A to B’ route, involving the turning of corners, a considerably higher system torque output would be needed, but such abilities are very much for the medium to long term.
Decades of development
The greater challenge in most respects is that of maintaining a safe distance behind any vehicle in front, or indeed of stopping (or going around) an unexpected obstacle. The genesis of such systems is found in the cruise control options that have been offered in many vehicles since the 1970s, and whose object is simply to maintain a steady speed regardless of circumstance.
The engineering of such systems has become much easier with the advent of ‘drive by wire’, the control signal being fed to the engine management unit rather than to a throttle transducer. It has also made easier the task of superimposing a signal, derived from a sensor measuring moving-vehicle separation or ‘headway’, to maintain a safe following interval even when the vehicle in front changes speed. Systems with this capability have become generically known as ‘intelligent cruise control’ (ICC).
In free-flowing main road traffic, ICC can normally be achieved with sufficient authority over engine output alone. However, any practical system must also allow for the vehicle in front stopping suddenly, or for the detection of a stationary obstacle that enters the forward-looking range of the sensor. This calls for the system also to be given authority over braking, and for the provision of a driver warning.
The sensors used in production ICC systems may be based on radar, laser or infra-red technology. Laser-based systems were an early favourite in Japan, but their performance falls away in adverse weather conditions and for some ‘targets’.
The European trend has been towards radar. A governing parameter is transmitter frequency: all other things being equal, the higher the frequency, the longer the effective range before atmospheric losses overcome signal strength.
After a long series of experiments and an amount of legal wrangling, automotive radar uses 77MHz or 24MHz. A typical 77MHz system has a range of 150 to 200m, depending on conditions; a 24MHz system is 50m or less. The payback is that 24MHz systems are significantly cheaper, partly because they use existing components from the telecommunications sector. In general, 77MHz systems are highly directional and have strong discrimination (in effect, the ability to distinguish the targets that matter from mere clutter). The 24MHz systems have a broader field of view; their role is more that of creating an electronic safety ‘cocoon’ and backing up the longer-range systems by triggering late-stage emergency braking when a target slips through the net, perhaps from a sideways or converging approach.
Legally, 77MHz has been accepted everywhere, but a question mark remains over 24MHz. It is an accepted standard in the USA, but in Europe is has been granted only temporary permission with limits on transmitted power.
However, many car models from different manufacturers are now offered with intelligent cruise control using 77MHz sensors, and these appear to work well, although further refinement of algorithms is desirable and possible – a crucial case, for example, is when another vehicle cuts across the beam in making a late exit from a motorway.
On the face of it, a 200m range is more than adequate, so long as decisions are made early enough: it is the stopping distance from 200km/hour (124mph) at 0.8g retardation, well within the dry-road deceleration ability of a modern car with ABS. That immediately raises the question not only of what to do on a wet road, but also of how soon maximum braking is applied, in other words the relationship between closing speed (after initial detection) and automatically applied braking effort. At the very least, the reaction time of the system is likely to be better than that of a human driver who realises there is a stationary obstacle 200m ahead on what he expected to be a free-flowing motorway. This is, of course, the worst case. Normally the vehicle ahead merely changes speed, and safe separation can be maintained by adjusting engine output alone.
In practice, until recently ICC systems were limited in their braking authority to around 0.3g, and indeed to a minimum speed of around 30km/hour. If they calculated that a stop before impact called for higher retardation, or for a full stop, they merely warned the driver to hit the brake pedal. These were the limitations of systems developed more with an eye to driver comfort in everyday driving than to providing protection in extreme circumstances.
Now some systems from Japanese manufacturers and recently from Mercedes with the latest iteration of its Distronic ICC have been made capable of applying maximum braking if the driver fails to react, and also reduce reaction time by pre-pressurising the braking circuit so that the mechanical reaction is nearly instantaneous.
The basic development of ICC (and lane departure) systems is very much the preserve of the large Tier One suppliers, who almost without exception have taken a close interest in this potentially large market with the promise of considerable safety benefits.
Among the Europeans, Bosch, Continental, Hella, Siemens and Valeo all have systems, and most are supplying them to vehicle-makers in one form or another. In Japan, Denso, Fujitsu and Hitachi are among the major players, and the large OEMs probably undertake more in-house work than their European counterparts. In the USA Delphi, TRW and Visteon all have major programmes.
The performance standard to be expected of an ‘extended ICC’ system capable of moving a vehicle along in stop-start urban traffic is more contentious. It is difficult to see how such a system could be applied other than to a vehicle with automatic or at least automated transmission. Otherwise there seems no reason why the concept should not be implemented, and the first production applications should arrive for the 2008 model year.
In the longer term, there is no question that a fully capable ‘autopilot’ should do what a capable driver does – plan ahead and anticipate. Thus a fully capable system should be able to take account of signals received from other vehicles and from roadside infrastructure.
A simple and much quoted example is automatic recognition of road signs: posted speed limits, advance warning of bends and so on. Even more helpful would be advance warning of trouble ahead, relayed back so that speed could be reduced in anticipation. An advanced autopilot might be capable of making lane-change decisions, after assuring itself it was safe to do so (‘blind spot’ driver warning systems are already offered in a number of upmarket models).
Automatic parking, an ability which has recently attracted much interest, would be another aspect of a fully integrated system. Ultimately, it is hard to imagine that precision GPS positioning would not join the suite of data sources drawn on by an autopilot.
The question of safety remains at the core of future advances. Again, the requirement is notably different from that of aviation. The immediate consequences of an errant autopilot are less serious in an airliner than in a road vehicle, which travels so much closer to fixed obstacles and in close company with other traffic. The classic aerospace approach of duplication (or triplication) and output comparison will not suffice.
The technique, already adopted in other areas, of ‘injecting’ frequent check-signals and making sure they are correctly responded to seems a far more likely option. The remaining requirement is for a clear set of legal standards, since it is surely only within such a framework that development teams – and company liability lawyers – will be happy to develop the concept of a vehicle autopilot to its logical and safety-enhancing conclusion.
|
| |
Author
Jeff Daniels
|
| |
| |
|
| Email this article |
| |
|
|
|
|