Features - Technical

DECEMBER 2, 1997

Colin McRae and other Control Systems

BY PETER WRIGHT

Two recent experiences, both as a passenger, have set me thinking about the driver as a control system.


Two recent experiences, both as a passenger, have set me thinking about the driver as a control system. With computer based control systems becoming more prevalent in many areas of motor racing, the interface between the driver and the racecar becomes ever more complex and important.

In August this year I experienced the National Automated Highway System Consortium (NAHSC) demonstration in San Diego, travelling as a passenger in a driverless vehicle along a 7.6 mile section of Interstate-15. The vehicle executed overtaking manoeuvres, obstacle avoidance and an emergency stop without human intervention. Perhaps more impressive was to be told that one of the demonstration vehicles had been driven from Washington DC to San Diego on normal highways, 98% of the time under automatic control.

Two weeks before the RAC Rally, I had the opposite experience - a ride beside Colin McRae in one of the 555 Subaru World Rally Team's Impezias, on a section of a Welsh forest test stage, as the guest of Prodrive. As we slithered sideways over the loose, wet surface, in and out of low cloud at what seemed ridiculous speeds, I got to thinking, as one does, about the driver as a control system (it was more calming than thinking about the pine trees flashing past the windscreen). At one point, when the car seemed to be at more than 90° to the direction of travel, I thought that its response had exceeded the driver's wishes. On querying this with Colin, he said "No, no, it was all under complete control".

The first experience, in San Diego, illustrated how a fully synthetic control system can, under certain controlled conditions, exceed the performance of the average human being. The objective is to make up for the deficiencies of the human when his/her performance is affected by tiredness, alcohol, drugs, stress and/or some physical distraction. It is part of the quest to combine the advantages of the car (privacy, individuality, flexibility, point-to-point access) with that of a train (passenger-density, control, professional operation).

The second, just north of Swansea, illustrated the ultimate in a human's ability to control a vehicle at the limit of its performance, in varying conditions. Being a World Rally car meant that the Subaru encompassed the most sophisticated control systems allowed in motorsport. It was more impressive than being driven round a race track, in spite of (or because of) the low g-levels. The track was narrow, with no space to scrub off speed in the event of a mistake, and lacked wide open safety features. It will be a very long time before a machine will be able to come near McRae's performance and a lot longer before I would have the confidence to sit beside it, in similar conditions.

Both the computer in the NAHSC vehicle and Colin McRae were doing the same job - controlling a vehicle to follow a path - and both used the same fundamental control system elements:

Sensors - Computer - Actuators

Although Colin McRae will probably be of more interest to Racecar Engineering readers, it is worth considering what the American engineers have had to develop to reach the stage where they can use an automatically driven vehicle on ordinary roads. The computer part of the control system is pretty straightforward - a Pentium PC does the job. Actuators to operate throttle, brakes, transmission, steering and signals are all to be found in any automobile maker's inventory. It is the sensors and the post-processing of their signals, plus the control algorithms that represent the achievement.

The key sensors, the ones that will control the cost of the system, are the forward, sideways and rearward-looking radars or lasars. Exploiting the Peace Dividend, the automobile industry has borrowed defence technology to produce devices that can do a better job than a human in his/her blind spots and in poor visibility; but they are expensive. Sophisticated post-processing is needed to interpret the radar returns and discard spurious warnings of objects. The state-of-the-art system works well.

However, radar cannot yet be used for path following. This is the area where the human eye and brain excel, when working properly. To emulate them, the engineers have used a simple low cost (less an $100) miniature TV camera and very sophisticated image processing. The task is to identify the edges of the lane in which the vehicle is travelling. This is reasonably straightforward if there are clear lane markers, but the system demonstrated was even able to pick up on the oil marks down the centre of each lane. It probably wouldn't work in virgin snow!

The control algorithms primary objectives are to maintain the vehicle in the desired position in the lane at the desired speed, and to change lanes when the situation requires it: just like wheels on rails. Speed control depends upon a number of inputs, including identifying the presence of other vehicles and data sent to it by similarly equipped vehicles and road side or central services.

It is an impressive system, and when the cost-benefits have been fully calculated and debated, in terms of potential savings in health costs and congestion losses, it, or something similar, will probably transform the traffic system. Impressive as it is however, it only works for vehicles operating well below the tyre adhesion limit, on prescribed road systems. The difference in its performance and that of a top rally driver at the limit on a special stage, serves to illustrate how the driver himself works as a control system.

When a rally driver reconnoitres a stage, driving it at low speed, he performs a similar path following task to the NAHSC vehicle. He will know the response of the vehicle having programmed his brain with its low speed characteristics. His eyes perform the TV camera and radar roles and his arms and legs are the actuators, inputting the control commands. These will be almost open-loop if he is driving slowly enough, as the response of the car in following the path is unlikely to require correction. He will also be storing away information about the path details, with the help of his co-driver/navigator and the latter's note pad. He will not need this information until he sets out to drive the car at the limiting speed. The moment he attempts to do this, he will transform himself into a much more complex system - a combination of open-loop and closed-loop, with sophisticated sensors for each.

To drive a given corner, the first stage is to estimate the maximum speed at which it can be taken, how hard it is possible to slow down from the current speed and hence where to start braking. The process is:

Figure:1

The human being does not really have any speed sensors, at least not any adapted to travelling in wheeled vehicles. When running, we know the length of stride and the frequency, and can estimate speed. We also use the noise of air rushing past our ears to help estimate speed, on a horse or bicycle for instance. Sit in a closed car and we need a speedometer. However when driving at the limit, the eyes are too busy to re-focus on an instrument, and so we take cues from the rate at which the scenery flashes past in our peripheral vision, plus the noise/vibration frequencies from the engine and wind noise to make a speed estimate - note how one always drives faster in a quiet car after becoming used to a noisy one, and vice versa. Experience of the conditions and the car count for a lot in determining the accuracy of the estimate.

Assessing an approaching corner is one of the great rallying skills. The co-driver can give a brief description and indicate which gear it should be taken in. This information primarily helps the driver recall his memory of the corner, but unless he is very familiar with the stage, it is the eyes that must gain all the necessary information about road surface, cambers, ditches, bumps, obstacles and corner radius to determine the optimum line and speed. Circuit racers do not have this problem, they learn each corner and pre-programme themselves with details of braking points, lines and speed. Only when the conditions change or there are other cars to pass, does the workload increase to that of a rally driver. See what happens when it suddenly rains - one or two drivers adapt immediately to the changing conditions and gain seconds over those who have to re-learn the circuit.

One essential piece of information the driver requires is grip level. I asked Colin how he determined braking points and he pointed out that he dabbed the brakes on the straights to calibrate the surface. After hearing this I concentrated on him doing just that, although almost no speed was lost in the process. The g-level is sensed by the semi-circular canals in the ears and by the harness loads, transmitted to the brain along with information about how hard the brakes have been applied, together giving an estimate of grip level available at that point. This information has to be combined with visual cues about the approaching corner.

The initial brake application is open-loop, based on experience. From this point on the driver goes into closed-loop control of the car. A circuit racecar can be driven at the limit virtually open-loop. I have seen it in data and one can often see it on the overhead TV camera output from a Formula 1 car. If the driver knows the circuit and car well enough, and has good confidence in that knowledge, he will input brake, steering, and throttle with almost no corrections, to take the car to but not over the limit of grip. Even though McRae had driven up and down the 4 kilometre test stage all day, he definitely was not driving open loop (the steering wheel was a blur much of the time). The conditions in rallying are too variable and the margin for error too low.

What does he do? Let us consider what he is trying to achieve in a 90° corner, for simplicity's sake.

Figure:2.

If he approaches the corner at say 100 mph and we define this direction of travel as the X-direction, then initially:

Vx=100, Vy=0

And the car is pointing along the X-axis.

After the corner, when the car is again doing 100 mph,

Vx=0, Vy=100

and the car is pointing at 90° to the X-axis.

The driver must control the car to follow the desired line around the corner, and end up with it pointing in the right direction, all conducted at the greatest possible speed. It is a matter of getting the X and Y co-ordinates of the car, along with the Yaw angle, b, correct at each instant. To do so he must sense and control the differentials (the calculus variety, not the mechanical ones) and integrals of X, Y and b i.e. accelerations and velocities as well as position. Sensing them is once again the key. X, Y and b are determined solely by the eyes and we can process these images quite quickly, so there are only small lags in sensing the position and attitude of the car. Velocities are more difficult, as discussed above. In particular Yaw velocity is a very unnatural requirement. I do not believe we can sense Yaw velocity with our eyes shut. It is the rate at which the scenery passes across the field of view that gives us a sense of rotational velocity. The processing of this visual information by the brain (performing continuous differentiation) is quite slow and has to take into account how far away the scenery is. If the head is shaken about, i.e. the eyes themselves are moving relative to the car, the noise-to-signal ratio prevents fine resolution of Yaw velocity.

Yaw velocity is the prime signal in controlling the cornering performance of the car. As the car travels around the corner the Yaw angle steadily increases until the car points in the new direction. The Yaw rate depends on the radius of the corner and the speed. Super-imposed on the Yaw angle is the slip angle of the car. Any change in Yaw angle, over and above that predicted by the normal passage of the car around the corner, signals a change in slip angle which the driver uses to control the lateral force. He must know the Yaw rate and Yaw angle all the time.

This is particularly true for rally drivers who use the lateral force, generated over a wide range of slip angles on gravel and snow tyres, interchangeably with the longitudinal force generated under braking and 4-WD traction. They throw the car sideways way before the corner, as part of the braking stage, and use power to accelerate the car in the new, corner exit direction.

The acceleration terms in the X, Y and b senses, are less important. The human body, through its acceleration sensors - the semi-circular canals and forces exerted on the body by seat and straps- is not very good at measuring absolute acceleration. What it is much better at doing is sensing small rates of change in acceleration, sometimes known as "jerk". The longitudinal and lateral accelerations on a car are steady through most of the corner, but any change will signal a loss of grip, an exceeding of the limit of the tyres, or a disturbance of the car. The "jerk" signal tends to be confused by "noise" generated by the motion of the car over bumps. One of the best features of an Active Suspension car is the stable and smooth platform, almost free of roll and roll-rock, from which the driver can control the car with the minimum of signal noise.

What the driver wants to know is any change in what is happening at the tyre contact patches, and he needs to know it with the minimum time lag - one essential of any good control system. A rally car has a natural frequency under 2 Hz - dictated by its inertia, the compliance in suspension and tyres, and its damping characteristics. By the time the driver has received the motion signals that something has changed, processed them and input a control correction, it may well be too late. Any means of conveying the essential information more quickly, is invaluable. The stiffest and lowest inertia connection between a tyre contact patch and the driver is the steering system. Providing it is free of stiction and compliance, and the geometry is correct, it will send signals of small changes in tyre self-aligning torque straight to his hands. This information can arrive with less than half the lag of other information and is in fact the prime sensing system for measuring the front axle grip - see Figure:3.

Figure:3.

For the rear axle, he has to rely on any change in Yaw acceleration to indicate a change in grip. He may also use changes in lateral acceleration, depending where he sits relative to the instantaneous centre of rotation of the car.

Thus the driver receives signals that give him the essential information he needs about position, speed and available grip, that enable him to adjust closed-loop, the brakes, throttle and steering, to gain the response from the car that he wants. How much feed-back he needs and how hard he has to work adjusting the response of the car will depend upon how accurate the initial, open-loop brake, steer and throttle inputs were. Which brings us to the subject of confidenceÉ.

It is a fundamental of both aircraft and racecar dynamics that an unstable vehicle will be the quickest to respond to inputs. However, an unstable vehicle will not give confidence to its pilot or driver. To overcome this problem, the aviation world has put a computer between the pilot and unstable aircraft, using sensors to detect the dynamic motion. Thus the pilot can make his demand and the computer will deliver, via the control surfaces, the maximum aircraft response that matches that demand, without need for correction by the pilot. Before computers found their way into aircraft control systems, the best approach was to provide the aircraft with good stability and equip it with powerful control surfaces. This is one of the keys to the Spitfire's legendary appeal to pilots, and is still the approach used in competition aerobatic aircraft (where computers, i.e. "Pilot Aids", are inappropriate) such as the latest Sukhois and Yaks.

The principles for a racecar are the same, but there is an important difference: on a conventional aircraft, the force generators (wings) are separated from the moment generators (tail plane/elevators). Designed correctly, the moments needed for stability and control will still be available when the wing gives up at it's limit i.e. stalls. On a car, the force and moment generators (tyres) are one and the same. At the limit they all tend to give up roughly at the same time - if the front axle gives up first, there is no control; if the rear axle first, there is no stability.

A top racing driver may be able to live with a marginally stable car. Michael Schumacher can, but his team mates tend to find it more difficult - they lose confidence in the car. The variable and unexpected conditions on a rally stage demand a stable and predictable car to give the driver confidence. The prime task for the engineers is then to provide by what ever means is possible, the maximum control of Yaw angle and Yaw rate. Sometimes their solution, particularly if it includes computer-control, clashes with the driver as a control system.

Rally drivers use a number of methods to persuade their cars to yaw. Initially turning the wrong way and then putting in the right steer angle causes the car to build up Yaw velocity and enables it to overshoot the stable slip angle. Once the steered wheels exceed their effective limit (note how much front wheel steer angle is used, as seen on TV, without discernible car response), torque distribution, brakes, power and differential characteristics, are all brought into play to provide a turning moment about the Centre of Gravity. Not only is the relative cornering power of each axle modified by applying power, but across-the-car torque distribution is used (see Differentials V..N....). The engineers must provide engine response, differential characteristics and a braking system to give the driver the control he needs.

So far as is common knowledge, most of these control systems are open-loop, except the engine. In particular, the differential controls set up the internal friction according to the conditions of power, speed etc at a given moment. They do not close the loop around the response of the car. Full closed-loop control systems that take over control of the yawing of the car have not found favour in racing. The frequency response of a human (around 5 Hz) compares very favourably with a car (Formula 1 cars: 4-5 Hz; rally cars: probably 1-2 Hz on gravel, 2+Hz on tarmac) and drivers have not responded well to systems that try to do what he is doing at the same time.

Such systems are ideal for road cars, being much faster than a soporific driver, or one in the middle of a phone call. The stability and control systems appearing on up-market road cars are just that. Fortunately for motorsport, I do not believe such systems will be of benefit to the likes of Colin McRae, and we shall be able to continue to be able to enjoy the visual impact of high performance control systems at work, for some time to come.