没有合适的资源?快使用搜索试试~ 我知道了~
温馨提示
试读
34页
这篇文章对人控制车进行研究,是从事无人车控制必看的文章! This paper examines the role of the human driver as the primary control element within the traditional driver-vehicle system. Lateral and longitudinal control tasks such as path-following, obstacle avoidance, and headway control are examples of steering and braking activities performed by the human driver. Physical limitations as well as various attributes that make the human driver unique and help to characterize human control behavior are described. Example driver models containing such traits and that are commonly used to predict the performance of the combined driver-vehicle system in lateral and longitudinal control tasks are identified.
资源推荐
资源详情
资源评论
Vehicle System Dynamics 0042-3114/03/4001–3-101$16.00
2003, Vol. 40, Nos. 1–3, pp. 101–134
#
Swets & Zeitlinger
Understanding and Modeling the Human Driver
CHARLES C. MACADAM
1
SUMMARY
This paper examines the role of the human driver as the primary control element within the traditional
driver-vehicle system. Lateral and longitudinal control tasks such as path-following, obstacle avoidance,
and headway control are examples of steering and braking activities performed by the human driver.
Physical limitations as well as various attributes that make the human driver unique and help to characterize
human control behavior are described. Example driver models containing such traits and that are commonly
used to predict the performance of the combined driver-vehicle system in lateral and longitudinal control
tasks are identified.
1. INTRODUCTION
It should first be noted that the general topics of (1) understanding human drivers and
(2) modeling their behavior, are quite broad in scope, either taken alone or together. In
order to bound this paper to a somewhat more manageable scale, the emphasis here is
on the control aspect of the human driver and its subsequent conceptual or computer-
based modeling. That is to say, many behavioral aspects of driving – less related to
common control task activities such as path-following, obstacle avoidance, or
headway control – are not being addressed. Such behavioral activities might include
items like driver distraction, side-tasking, or driver impairments that are more in the
realm of human factors interests. Rather, the focus here is on the steering or braking/
acceleration control behavior of human drivers commonly observed to be present for
both lateral and longitudinal control of the vehicle.
The principal roots of driver modeling as it relates to these control activities extend
back to the early years of human-machine and aircraft pilot studies [1–14]. Those
studies helped to reveal various properties unique to, or characteristic of, the human as
a controller of dynamical plants [1, 4, 8, 14, 15] and certain vehicles [6, 16–20]. For
example, in 1961 Ornstein [8] proposed the following transfer function, H(s), model
of the human operator for pursuit-type manual tracking tasks,
Hðs Þ ¼ ða
1
$ s þ a
0
Þ$e
&s!
=ðb
2
$ s
2
þ b
1
$ s þ b
0
Þð1Þ
1
Address correspondence to: C.C. MacAdam, The University of Michigan, Transportation Research
Institute, 2901 Baxter Road, Ann Arbor, MI 48109 USA. Tel.: (734) 936 1062; E-mail: cmacadam@
umich.edu
Downloaded By: [University of Michigan] At: 21:34 22 February 2010
and noted that the numerator coefficient a
1
associated with the velocity component, or
anticipatory behavior of the human operator, was adaptive to changing plant
dynamics and methods of visual presentation. In effect, this was an early observation
of the adaptive nature of a human when interacting with different or changing plant
dynamics and the use of prediction by the human operator. The parameter t was an
effective transport time delay.
More specific studies aimed at understanding the human as an auto mobile driver
[16, 19–31] followed from or paralleled this type of manual control work. One
example is a series of publications by Rashevsky in the late 1950s and early 1960s that
addressed the topic of ‘‘Mathematical Biology of Automobile Driving’’ [25] wherein
the basic model of the driver as a steering controller was treated in a purely geometric/
kinematic manner. The proposed algorithm for steering an automobile of designated
width and length along a straight road of specified width amounted to issuing
instantaneous steering corrections to the vehicle whenever it approached to within a
specified margin of either lane edge. Although no vehicle dynam ics were considered
within this mathematical treatment, driver anticipation and driver delay properties
were include d as key parameters. Rashevsky interestingly observes in one work [32],
To sum up, we see that the combination of human parameters and of mechanical
parameters enter into the process of driving in a manner which does not permit
their clear-cut separation. The car and the driver form, in a sense, an individuum.
And in another work [33],
The car and driver constitute a complex feedback system. The behavior of the car
results in certain reactions by the driver. Inversely, the behavior of the driver affects
the behavior of the car. This ‘man-machine’ system cannot, in many instances, be
separated into the purely ‘mechanical’ and the purely ‘human’ components. The
system must be treated as a whole.
So we see that the idea of treating the driver and vehicle together as a combined
system was recognized as important by researchers, absent even dynamical
considerations of the vehicle response.
More complete treatments of the driver modeling effort and associated
measurements of driver-vehicle system responses were also emerging in this same
time frame, particularly with the increased use of computers and driving simulators.
Example works that studied driver control behavior and accounted for the dynamics
of the vehicle include studies by Wierwille [19], McRuer [22], Weir [18, 29–31],
Kondo [17], Yoshimoto [20] as well as many others [16, 26–28, 34 –37]. Key findings
from these studies and others that fol lowed are discussed further in subsequent
sections of the paper. Other survey papers related to the topic of driver control
behavior and modeling are found as well in such works as Good [38], Reid [39], Guo
[40], and Peng [41].
102
C.C. MACADAM
Downloaded By: [University of Michigan] At: 21:34 22 February 2010
In many cases, motivation for understanding and modeling the driver may be
primarily related to achieving improvements in vehicle design per se, insofar as an
interactive, or cause and effect, relationship is recognized between the vehicle and the
driver [28, 38, 42–51].
2. CHARACTERISTICS OF THE HUMAN DRIVER
This section of the paper describes various human characteristics, both in terms of
physical limitations as well as certain attributes, that should be incorporated into any
serious effort aimed at modeling the control behavior of the human driver. Various
example works – both historical as well as current – document many of the pertinent
aspects of human control [15, 52–60]. The first subsection primarily addresses
physical limitations of human sensory and physiological abilities and related driving
cue issues [61]. This is followed by discussion of various human attributes that
include such items as compensatory control behavior exhibited by drivers during
regulation tasks, driver preview utilization, and adaptation capabilities of drivers
when confronted with altered vehicle dynamics and/or changing operating conditions.
2.1. Physical Limitations
2.1.1. Human Time Delay and Threshold Limitations
There are certain basic properties of humans that are routinely observed by
experimentalists when studying human-machine interactions. Humans are not
‘‘linear’’ elements. They exhibit time delays in reacting to stimuli. Sensed
information (visual stimuli, motion cues, etc.) must also exceed certain thresholds
prior to being detected. Other general limitations are:
' required processing times for sensed information
' information transmission time
' cognitive requireme nts to anticipate or predict ahead
' perceptions of higher derivative (or rate) information
Pure time delays are seen to be more harmful to human-vehicle system
performance than ‘‘exponential’’ or dynamic lags as associated with first or second
order system characteristics [13, 62]. Laboratory experiments involving tracking tasks
show that time delays greater than 40 ms produce degradations in performance for
zero-order systems (i.e., simple positional control tasks) [13]. The detrimental
influence of pure time delays on system stability was also demonstrated [9].
Simple reaction-time experiments (the time from stimulus to response with
anticipation) show that visual response times under near ideal conditions are about
180 ms. Comparable auditory and tactile response times are about 140 ms [59, 60, 63].
UNDERSTANDING AND MODELING THE HUMAN DRIVER 103
Downloaded By: [University of Michigan] At: 21:34 22 February 2010
It was also demonstrated that auditory delays increase to 300–400 ms as sound
amplitudes approach the threshold detection levels [60]. Others have observed that the
duration of the stimulus plays a role in reaction-times [64]. Consequently, reaction-
time characteristics can display nonlinear properties insofar as they depend upon the
amplitude as well as the duration of sensed information.
Some example threshold values for humans are listed below for several sensory
channels (assuming long latency detection times in some cases) [52, 53]. Many of
these threshold values depend upon exposure time, location, etcetera, and can be
considerably more complicated than indicated by these example values:
0.005 g linear accelerations vestibular saccule
0.1 deg/s/s rotational accelerations vestibular semicircular canals
5 mg (von Fry hair method) tactile – female face
355 mg (von Fry hair method) tactile – male large toe
0.0002 dynes/cm/cm auditory
(0.63 dynes/cm/cm is ordinary conversation)
2.1.2. Visual Characteristics
It is known that velocity and position information can be extracted separately from
a scene by the human vision system [65]. The jump-like saccadic response of the
eyeball is primarily triggered by positional errors; the smoother pursuit mode of the
eyeball is activated for constant velocity tracking of objects. However, the two
modes work toge ther yet independently of one another [52, 53]. There is evidence
that cells within the human visual system are directly responsive to velocity
[66–69].
The perception of velocity is known to impose costs of increased time delays in the
range of 30–200 ms [5, 70, 71]. Limited abilities to perceive acceleration via the
visual channel are observed except in cases where velocity-differencing operations
occur by humans for slowly moving targets [52, 72].
Velocity information presented to humans in displays as ‘‘positions’’ (e.g., a
speedometer gauge) show improvement in performance for tracking tasks, over cases
in which velocity information is presented as a ‘‘velocity display’’ (e.g., nulling of a
rotating disk display) [73]. Velocity information obtained from the peripheral visual
areas is generally found to be valuable, particularly as redundant information to
positional information obtained from the central foveal vision areas [74–76]. It is also
observed that basic peripheral velocity information should be compatible with the
expectations of humans [76].
The ability of humans to look ahead and preview information helps to reduce some
of the time delay limitations indicated above [77]. In other cases where preview may
not be available, such as driving and responding to wind gusts, higher order predictive
104
C.C. MACADAM
Downloaded By: [University of Michigan] At: 21:34 22 February 2010
requirements of humans come into play in order to self-generate rate information for
vehicle control purposes.
2.1.3. Motion Influences
Much of the higher derivative information used by humans for control purposes is
obtained from the vestibular (inner ear) and kinesthetic (body distributed) channels.
Simulator studies show that the presence of accurate motion information to
supplement visual information generally leads to superior flying and driving
performance over comparable conditions when no motion is present [3, 6, 73, 78–84].
Studies such as [6, 85] show that motion effects help to reduce uncertainties in
observed responses and provide for enhanced prediction and gain com pensation by
the human controller. The importance of accurate and reliable vestibular cue
information is noted in simulator studies such as [6]. Cue fidelity issues have also
been noted with regard to pilot training and successful transfer to flight conditions.
Experience or skill level may also play a role as to cue preferences during simple
driving tasks as suggested in [86].
2.1.4. Auditory Information
The use of auditory information is generally seen to be most beneficial when acting as
a supplementary cue within a multi-channel environment [87, 88]. For tracking
control tasks, it is noted in some studies that human time delay characteristics were as
short for auditory as for visual channels [88].
Generally, the use of auditory information as redundant and reinforcing infor-
mation is seen as helpful for improving system performance. Auditory information,
though, is most useful under high workload conditions as redundant information
supplementing the visu al channel [53].
2.1.5. Tactile and Haptic Information
Tactile and haptic cue information is normally conveyed to the driver through the
steering wheel and throttle/accelerator pedals. A certain portion of information is also
available through such channels by sensing small skin surface vibrations or
circulating wind, etc. Steering wheel torque information can be particularly useful
to drivers for detecting sudden changes in tire/roa d friction as well as anticipating
control responses for roadway disturbances and wind gusts. Example ideas being
advanced in this area are the so called haptic steering wheel device [89] and certain
electric steering systems [90].
2.1.6. Ranking of Sensory Cue Influences for Human Drivers
General review of the overall literature leads to a clear impression that visual aspects
of driving are of the highest significance. Quotes referring to the driving process as
depending upon 90% of visu al information are not uncommon [91–93]. There is also
UNDERSTANDING AND MODELING THE HUMAN DRIVER 105
Downloaded By: [University of Michigan] At: 21:34 22 February 2010
剩余33页未读,继续阅读
资源评论
zfy1111
- 粉丝: 0
- 资源: 10
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功