A Novel Metric for Evaluating Human-Robot Navigation Performance

Article (PDF Available) · January 2006with 65 Reads
Cite this publication
Abstract
It is frequently desirable to quantitatively evaluate perfor-mance in human-robot interaction. In this research, we in-troduce a novel use of fractal based performance measures to evaluate robot navigation. It can be difficult to use simple measurements such as path length and task timing due to large amounts of individual variability along these metrics. This shortcoming makes these measures particularly unsuit-able for evaluating human/robot navigation performance in a complex environment with multiple tasks. Herein we out-line a class of novel metrics for evaluating navigation perfor-mance using fractal geometry. This class of space and time invariant metrics allows us to better analyze human-robot exploration, interpret valuable behavioral information for analysis of movement, measure the handler/robot's search efficiency, path tortuosity, and overall space utilization in relation to handler goals and overall characteristics of the environment. We present techniques for computation and empirical results from our own navigation interface studies.
Figures - uploaded by Flip Phillips
Author content
All content in this area was uploaded by Flip Phillips
Content may be subject to copyright.
A Novel Metric for Evaluating Human-Robot Navigation
Performance
Flip Phillips
Skidmore College
Psychology & Neuroscience
Vision Laboratories
Saratoga Springs, New York 12866
flip@skidmore.edu
Martin Voshell
The Ohio State University
College of Engineering
Cognitive Systems Engineering Laboratory
Columbus, Ohio 43210
voshell.2@osu.edu
ABSTRACT
It is frequently desirable to quantitatively evaluate perfor-
mance in human-robot interaction. In this research, we in-
troduce a novel use of fractal based performance measures to
evaluate robot navigation. It can be difficult to use simple
measurements such as path length and task timing due to
large amounts of individual variability along these metrics.
This shortcoming makes these measures particularly unsuit-
able for evaluating human/robot navigation performance in
a complex environment with multiple tasks. Herein we out-
line a class of novel metrics for evaluating navigation perfor-
mance using fractal geometry. This class of space and time
invariant metrics allows us to better analyze human-robot
exploration, interpret valuable behavioral information for
analysis of movement, measure the handler/robot’s search
efficiency, path tortuosity, and overall space utilization in
relation to handler goals and overall characteristics of the
environment. We present techniques for computation and
empirical results from our own navigation interface studies.
Categories and Subject Descriptors
H.4 [Information Systems Applications]: Miscellaneous;
D.2.8 [Software Engineering]: Metrics—complexity mea-
sures, performance measures
General Terms
Fractal dimension, Tortuosity, Human-Robot Coordination
1. INTRODUCTION
Consider the following scenario— a teleoperated robot nav-
igates through a collapsed building and encounters a large,
fallen beam. In a search and rescue mission the operator
faces multiple task-critical decisions— for example, can the
robot fit through a small opening, is the robot able to travel
over the obstacle, or should they simply backtrack and find
another way to the target area? Several constraints frame
decisions of these sorts. For example, resources are at a
Human Robot Interaction ’06 Salt Lake City, Utah USA
premium with respect to onboard power, time is critical in
rescue missions, and cognitive mapping and machine oper-
ation performance varies widely between operators. When
evaluating mission performance, the complexity of the afore-
mentioned constraints suggests that relatively simple met-
rics may not adequately capture the nature of these noisey
relationships on performance.
When exploring a disaster environment with Urban Search
and Rescue (USAR) robots or when monitoring autonomous
military vehicles, energy conservation and battery life are
at a premium and human-robot handlers and supervisors
need meaningful ways to measure and assess the perfor-
mance of the deployed platforms. The human operator (or
the onboard autonomy) is tasked with controlling and reg-
ulating a remote robot platform to fulfill stakeholder and
handler goals in a dynamic and changing world. The result-
ing human-robot behaviors displayed stem from this agent-
environment interaction. In a Gibsonian sense, we are in-
vestigating the constraints that perceptually mediated inter-
faces for remote robot control have on human perception ev-
ident in goal-driven spatial exploration. The interfaces and
hardware used are the constraints placed on a human opera-
tor’s (and therefore the robot’s) actions and limit or expand
the functional coupling between the human, the robot, and
the environment. Any mediated problem of this nature fun-
damentally becomes an issue of coordinating these various
facets to perform a task in an environment. Performance
measures are needed that can better capture these relation-
ships and how they change when introducing new operator
interfaces, new levels of autonomy, and new levels of super-
visory control to the human-robot system.
Teleoperation has the potential to create many perceptual
ambiguities and, with differing levels of visual access, oper-
ators tend to employ a variety of different navigation strate-
gies. People learn and adapt and change their search strate-
gies constantly. When researchers and engineers introduce
new system component designs, gauging the responses are
quite complex and simple outcome measures such as target
‘hits’ and completion time are not likely to be very effec-
tive. Evaluating navigation and exploration in such settings
presents many challenges when it comes to analyzing per-
formance data. This holds equally true for measuring au-
tonomous robot behavior. Similar to Smithers’ [14] lament
over the state of quantitative assessment of computational
algorithms, there are still no meaningful quantitative or
DRAFT COPY- Flip Phillips, Martin Voshell, 2006
DRAFT COPY- DO NOT REPRODUCE
Figure 1: Paths shown with increasing fractal dimensions. The top path is fairly direct and, as such, has a
very low fractal dimension. The bottom path has several areas that have been traveled over multiple times.
Using a strict time or distance based measure would overestimate the amount of the space explored in this
case. A fractal measure is more conservative.
qualitative indications of how well a robot is carrying out
its navigation tasks, how well a robot avoids obstacles, or
generally how well a robot keeps from getting stuck.
In the robotics and human factors literature, analysis met-
rics have a difficult time capturing the intricacy and com-
plexity involved in the domain of practice, especially when
it comes to bridging metric representations with cognitive
work from each field. Fong et al. [5] emphasize the need for
a standardized framework of metrics that span across the
application space between humans, robots, and the human-
robot systems. The most significant problem they cite in
utilizing common metrics lay in the fundamental issue that
that human-robot teams simply are utilized in many diverse
applications.
In our previous research designing interfaces for teleoperated
robots in USAR settings [15, 16], one of the central compo-
nents of analysis was navigation. Investigators looking at
such navigation performance data tend to employ multiple
combinations of measures usually within three dimensional
space and time. Chronometric and tally measures collect
task completion times, rest periods, and mean number of
stops. Other metrics tend to look at Cartesian properties
of the exploration, looking at distance traveled, distance
backtracked, number of turns, and number of stops. Many
measurments are entirely task specific, keeping track of tar-
gets located and identified, missiles fired, mines defused, etc.
Others seek more descriptive measures such as smoothness
of trajectory, mean absolute error in pointing direction, and
number of collisions. Clearly, there are many potential met-
rics available, however none of these tell a complete story.
Such traditional measures for analyzing the robot handlers
performance in a complex environment with multiple tasks
were not very informative and it proved very difficult to
pull out different individual search strategies in what can be
seemingly random exploration behavior with high levels of
interoperator variance. Simple first order tabulations tend
not to tell much about complex processes nor capture the
intricacies and problems inherent in search and rescue in
remote exploration. We propose using scale invariant fractal
techniques to be able to help meaningfully understand such
complex performance.
Different types of relational measures were needed to be able
to look at goal-oriented paths in relation to environmental
characteristics and challenges. To this end, we adapt frac-
tal measures from their uses in experimental biology, phys-
iology, and entomology. Whereas traditional quantification
and comparison measures focus on what sources are produc-
ing observed variation, fractal describes the variation itself.
Deviations in the fractal dimension of the robot/handler’s
goal directed path through a complex environment in rela-
tion to various ambiguities and obstacles provide a metric
that captures a very rich set of descriptive behaviors allow-
ing performance evaluation across many applications.
2. FRACTALS AS PERFORMANCE MET-
RICS
How can we gauge how well a human/robot system is cur-
rently performing? In many ways search and rescue is anal-
ogous to a foraging task in an unknown environment. For
any animal, human, or robot, a complex environment struc-
ture constrains movement. When functioning under these
constraints, searching for distributed information in a given
area becomes challenging. In many biological settings track-
ing exploratory behavior is quite complex and seemingly
random. What could we learn from these and could it be
applied to the data we were finding in simulated teleopera-
tion?
Dicke and Burrough [4] first utilized fractal metrics to ex-
plain spontaneous locomotion data in insects with respect
DRAFT COPY- Flip Phillips, Martin Voshell, 2006
Figure 2: The images on the right show god’s-eye-views of two paths taken by subjects in two different
interface configurations. The top shows a multi-camera ’folded’ interface and its respective fractal dimension,
D= 1.03 and the bottom shows a traditional single camera interface with D= 1.18. The second human /
robot team faced significant trouble navigating around the fallen beam in the illustration on the left since
they could not adequately determine if they could fit the robot through the available opening.
to their position. Initially, this motion appeared to be ran-
dom but, when analyzed with a sensitive nonlinear measure,
turns out to be structured in a more complex domain. In
these measures, the fractal dimension, D, as proposed by
Mandelbrot [8], provides a measure of this hidden regular-
ity.
Briefly, Dprovides a measure of path complexity that is
totally scale- and time-invariant. Imagine traveling on a
path between two discrete locations, AB, on a two-
dimensional plane P. When traveling in a straight line be-
tween them D= 1.0, the same as its Euclidian counterpart.
If, on the other hand, you were to systematically explore
every location on Pin an effort to find Byou will cover the
two-dimensional area of Pand thus, D= 2. Totally ran-
dom (i.e. Brownian) motion through the space results in a
D= 1.5. This suggests that, when applied to position of
a robot vehicle, this measure provides a useful indication of
degree of exploration independent of the time or path length
taken (See Figure 1).
In the natural world, ecological objects tend to be highly
fragmented and irregular and individual behavioral differ-
ences are often large enough to make quantifying such be-
havior quite difficult. In the animal behavior research such
individual variation may provide valuable information about
what components are leading toward specific behavioral se-
lections [11, 6]. Many species and many behaviors have
been described in 2-D and 3-D as fractal. From the many
scales of variation inherent in the spontaneous locomotion
trails of spider mite spatial exploration described by Dicke
and Burrough [4], to foraging patterns of minnows [1], hawk
prey vigilance [7] and even in analyzing social behavior se-
quences in chimpanzees [2], fractal measures provide a mean-
ingful and therefore useful metric. When describing goldfish
swimming, Neumister explains that fish often swim in highly
organized spatial patterns consisting of coordinated motor
activity where individual activity is often explained with
socio-biological descriptors with little concern for those in-
dividual differences [11]. Persistence (repeating squences)
and redundancy (uncertainty to length) in fish swimming
are two such nonlinear operationally defined patterns that
Neumeister used different fractal approaches to quantify.
Human/robot pairs must explore highly irregular terrain,
they might have to go around corners, over or under de-
bris and rubble, and the optimal path is seldom a straight
line. From our own work [15, 16] we have seen there are
large amounts of interoperator variability in path explo-
ration and Nehmzow and Walker [10] show that some au-
tonomous pathing exploration tends to exhibit determinis-
tic chaos. Quantifying such complex and seemingly random
data while taking into account behavior, task, and locomo-
tion, fractal measures serve as a promising nonlinear mea-
sure to characterize such human-robot exploration.
2.1 Applications
An application of interest to us concerns assessing the qual-
ity of navigation performance in order to compare several
teleoperation interfaces in the USAR domain [15, 16] . A
simulation of a teleoperated robot-assisted search and res-
cue mission was carried out in an urban environment (based
DRAFT COPY- Flip Phillips, Martin Voshell, 2006
Figure 3: Exploring a disaster area results: path length and fractal D when navigating around a fallen beam.
In the first group, using the ”folded” interface, the path length and fractal dimension increase concomitantly.
In the ”single” trials the relationship between path length and fractal dimension is no longer systematic. In
the case of SM the path lengh is almost 2×the more spatially complex path of JG. This indicates that SM
spent a larger amount of time backtracking on the same path rather than exploring as JG did. Additionally,
the mean distance traveled is significantly longer in the single-camera interface condition, yet the fractal
dimension remains the same, suggesting that the amount of area explored overall remains constant regardless
of interface.
on [13]). The task scenario was based on interactions with
disaster specialists and mimicked the role of a structural
triage team assessing the general condition, type of occu-
pancy, and the collapse mechanism of a building disaster.
The robot-handler navigated a virtual robot through the
disaster area and had to gather as much information as pos-
sible about the incident site to both evaluate the need for
specific specialists as well as to ensure rescuer safety prior
to sending in a reconnaissance team. The disaster envi-
ronment was broken into specific transit and special stages.
Special stages were essentially probe areas and consisted of
particular perceptual ambiguities and cognitive challenges
for the robot-handler. Analysis was focused on the fractal
dimension, path length, and velocity in these special stages
to better understand the implications various interfaces had
on robot-handler environment exploration. In Figure 2, two
of the interface concepts we examined are illustrated — A
single small field of view camera versus an array of multiple
cameras.
As interface designers we are attempting to improve han-
dler/robot behavior by making display data meaningful. The
fractal analysis captures the relationship between the han-
dler and their understanding of the environment evident
through their locomotion in the environment relative to ob-
stacles, apertures, and other perceived affordances. The
fractal dimension provides an affordance-centered metric that
captures the efficiency of the space explored and is a rich in-
dication of how well an individual is performing with a given
interface. The initial findings from this work suggest that
when interfaces are catered to specific environmental condi-
tions, the operators’ robots have less tortuous paths.
During testing, the fractal dimension was used to character-
ize the tortuosity of the handler/robot paths. Our virtual
robot collected telemetry information in the form of x, y, z
position at the rate of 2Hz. Subsequent to each run we com-
puted a fractal Dfor the path’s spatial component using a
custom Mathematica [17] package. The computation was
performed using a plurality of methods as outlined in [9] to
check for convergent results. Our final method was a hybrid
of the the box method combined with the caliper method,
using multiple scales of interpolated analytic representations
of the path covered. In the biological literature, ‘tortuosity’
refers to how erratic and contorted an animal’s path is and
fractal models have been used to describe the shape of these
goal-directed paths in regard to how directional (D= 1.0)
or random they are (D1.5). Figure 3 show results for
six subjects spilt across the two different experimental con-
ditions — three subjects using an interface consisting of an
array of cameras and three subjects using a traditional sin-
gle camera view. Our results show that the mean distance
traveled is significantly longer in the single-camera interface
condition, yet the fractal dimension remains the same. sug-
gesting that the extent of overall exploration remains con-
stant regardless of interface, but the multi-camera interface
cuts down on revisiting the same locations.
Fractal analysis is not merely restricted to the spatial do-
main. Van Orden, Holden, and Turvey [12] suggest that
temporal properties of human performance also obey highly
non-linear rules. Seemingly random or noisy performance
in time-domain tasks reveal regularities when measured us-
ing fractal metrics. (See [3] for a survey of methods used
in physiological measurements.) Figure 4 shows the speed
of the robot throughout the course of a trial for two sub-
jects. A visual examination of the speed profile suggests a
rather noisy performance but again, the fractal results of
D= 1.16 for Subject A (on the left) and D= 1.33 for
Subject B (on the right) suggest that B’s performance with
respect to speed was less consistent than the performance
of A. Even though Subject B may have completed the task
in a shorter amount of time, the acceleration / deceleration
patterns may have required more energy utilization than
Subject A’s run. More to the point, this analysis can be
continued with higher derivatives of position as well, such
DRAFT COPY- Flip Phillips, Martin Voshell, 2006
100 200 300 400 500
50
100
150
200
time
(seconds)
time
(seconds)
speed
(units/sec)
Fractal D = 1.16 Fractal D = 1.33
0 50 100 150 200 250 300
50
100
150
200
speed
(units/sec)
Figure 4: Speed profiles for two Subjects, A (left) and B (right). Both performances are visually ’noisy’
but Subject B’s performance is significantly more-so according to the fractal metric. Even though Subject B
may have completed the task in a shorter amount of time, the acceleration / deceleration patterns may have
required more energy utilization than Subject A’s run.
as acceleration. Furthermore, this analysis could be done
directly on energy utilization, even providing warnings to
the operator as their utilization becomes more ‘fractal’ and
inefficient.
3. DISCUSSION
Assessing complex performance data in remote teleopera-
tion and observation is a difficult problem. Performance
tends to be extremely complicated and before we can assess
effective ways to deal with remote perception we have to
develop methods and performance metrics that capture the
interactions between the human, the robot, and the envi-
ronment to better evaluate human-robot team performance.
When designing new interfaces or testing new autonomous
navigation systems, it is imperative to utilize metrics that
capture how well a platform team is doing in spite of what
can appear to be highly variable and random data.
In this paper, we have proposed using fractal techniques
as quantitative estimates of overall human-robot navigation
behavior. We can look at these performance measures in
other spaces rather than strictly time, space, and we can
calculate a scale-invariant measure that is useful integration
of space-time. The fractal approach allows us to specify a di-
mensionality independent of scale and allows us to compare
and contrast it with other simpler scale dependent measures
to help elucidate the true nature of the operator/robot’s per-
formance. This analysis can be extended beyond the spatial
domain to temporal measurements such as speed and energy
utilization.
As onboard navigation improves, the robot handler role will
slowly translate to one of supervisory control as teleopera-
tion gives way to the need to coordinate many agents with
mixed levels of autonomy. At this point, supervisory con-
trol intervention and the ability to ascertain if a platform
is underperforming becomes critical. Integrating such per-
formance metrics as a robot’s instantaneous fractal dimen-
sion into a supervisor’s interface puts the robot performance
data into context and could give valuable cues to status at
a glance telling a supervisor when to intervene, or adjust
control. Fractal can be adapted across many scales- from
a small UGV to a USAR robot, a micro UAV to Predator
drone UAV, we have a means to start looking how to adapt
different classes of general robot behaviors at different spa-
tial scales. Our current experiments further examine the
efficiency and veridicality of mental representations built up
in exploration, as a function of the fractal dimension of the
exploration.
On a much larger level, such biological approaches have
many potentially useful implications toward ecological robot
design. By treating robot’s telemetry data in such a way,
valuable behavioral information can be interpreted for anal-
ysis of movement and overall environment utilization. Using
fractal to describe such goal-oriented behavior of human op-
erated robot movements in complex environments is a start-
ing point to try to integrate such measures into future op-
erator interfaces, as well as the control systems themselves.
4. ACKNOWLEDGMENTS
Prepared through collaborative participation in the Advanced
Decision Architectures Consortium sponsored by the U. S.
Army Research Laboratory under the Collaborative Tech-
nology Alliance Program, Cooperative Agreement DAAD19-
01-2-0009.
5. REFERENCES
[1] C. Alados and D. N. Weber. Lead effects on the
predictability of reproductive behavior in fathead
minnows (pimephales promelas): a mathematical
model. Environmental Toxicology and Chemistry,
18(10):2392–2399, 1999.
[2] C. L. Alados and M. A. Huffman. Fractal long-range
correlations in behavioral sequences of wild
chimpanzees: a non-invasive analytical tool for the
evaluation of health. Ethology, 106:105–116, 2000.
[3] J. B. Bassingthwaighte, L. S. Liebovitch, and B. J.
West. Fractal Physiology. Oxford University Press,
1994.
[4] R. R. Dicke and A. P. Burrough. Using fractal
dimensions for characterizing tortuosity of animal
trails. Physiological Entymology, 13:393–398, 1988.
[5] T. Fong, D. Kaber, M. Lewis, J. Scholtz, A. Schultz,
and A. Steinfeld. Common metrics for human-robot
DRAFT COPY- Flip Phillips, Martin Voshell, 2006
interaction. In IEEE Int’l. Conference on Intelligent
Robots and Systems, Sendai, Japan, 2004.
[6] V. Gotceitas and P. Colgan. Individual variation in
learning by foraging juvenile bluegill sunfish. Journal
of Comparative Psychology, 102(3):294–299, 1988.
[7] C. Loehle. Home range: a fractal approach. Landscape
ecology, 5(1):39–52, 1990.
[8] B. B. Mandelbrot. The Fractal Geometry of Nature.
W. H. Freeman, 1982.
[9] J. D. Murray. Mathematical Biology. Springer, 2002.
[10] U. Nehmzow and K. Walker. The behaviour of a
mobile robot is chaotic. AISB Journal, 1(4), 2003.
[11] H. Neumeister, C. J. Cellucci, P. E. Rapp, H. Korn,
and D. S. Faber. Dynamical analysis reveals
individuality of locomotion in goldfish. The Journal of
Experimental Biology, 207(697-708), 2004.
[12] G. C. V. Orden, J. G. Holden, and M. T. Turvey.
Self-organization of cognitive performance. Journal of
Experimental Psychology: General, 132(3):331–350,
2003.
[13] Pandora-Studios. Frag ops: A total UT 2004
conversion (http://www.frag-ops.com/), July 2005.
[14] T. Smithers. On quantitative performance measures of
robot behaviour. Robotics and Autonomous Systems,
15:107–133, 1995.
[15] M. G. Voshell. Overcoming the keyhole effect in
human-robot coordination. Master’s thesis, The Ohio
State University, Columbus, Ohio, 2005.
[16] M. G. Voshell, D. D. Woods, and F. Phillips.
Overcoming the keyhole effect in human-robot
coordination: Simulation and evaluation. In
Proceedings of the 49th Annual Meeting of the Human
Factors and Ergonomics Society, Orlando, FL, Sep.
26-30 2005. Human Factors and Ergonomics Society.
[17] S. Wolfram. Mathematica: a system for doing
mathematics by computer. Addison-Wesley, Redwood
City, Calif., 2nd edition, 1991.
DRAFT COPY- Flip Phillips, Martin Voshell, 2006
  • Article
    Full-text available
    This paper discusses a new real-time fractal path analysis (RTFPA) algorithm and its use in assessing the skill of robot operators. Twenty-five volunteers participated in an experiment to evaluate the use of the RTFPA algorithm as a metric for robot operator skill in a search task. The algorithm was used to estimate the fractal dimension of a path taken by a simulated, tele-operated Inuktun Extreme VGTV within a game-based training environment. The results show that within the sample population there exists a curvilinear relationship between fractal dimension and score. This relationship seems to indicate that when used along with score, fractal dimension can be used as a measure of an operator’s search strategy and ability to maintain situational awareness. An additional analysis confirms the results of prior work showing that there is no evidence of a relationship between fractal dimension and task completion time. Additionally, a detailed description and pseudo code for the RTFPA algorithm are presented and the accuracy of RTFPA is compared to existing fractal path analysis algorithms. KeywordsFractal dimension–Human-robot interaction–Teleoperation–Training–Video games
  • Article
    Full-text available
    Substantial automation will be needed to allow operators to control the large teams of robots envisioned for search and rescue, perimeter patrol, and a wide variety of military tasks. Both analysis and research point to navigation and path planning as prime candidates for automation. When operators are isolated from robot navigation, however, there may be loss of situation awareness (SA) and difficulties in monitoring robots for failures or abnormal behavior. Operator's navigational strategies are quite complex and extremely changeable at foraging tasks in unknown environment reflecting background knowledge and expectations about human and natural environments. These considerations are missing from automated path planning algorithms leading to differences in search patterns and exploration biases between human and automatically generated paths. Effectively integrating automated path planning into multirobot systems would require demonstrating that: 1-automated path planning performs as well as humans on measures such as area coverage and 2-use of automated path planning does not degrade performance of related human tasks such as finding and marking victims. In this paper we seek to compare the divergence between human manual control and autonomous path planning at an urban search and rescue (USAR) task using fractal analysis to characterize the paths generated by the two methods. Area coverage and human contributions to mixed-initiative planning are compared with fully automated path planning. Finally, the impact of automated planning on related victim identification and marking tasks is compared for automated paths and paths generated by previous participants.
  • Article
    Lead (Pb) has been shown to affect the behavior of a wide variety of vertebrates, including fish, amphibians, and mammals. This article re-examines previous data on the effect of short-term, sublethal levels of waterborne Pb on the reproductive behavior of fathead minnows (Pimephales promelas). Previous research has found that Pb decreased the time spent in displaying specific reproductive behaviors in male minnows. Because each activity performed within a sequence depends upon previous parts of the sequence, the reproductive behavior of fish is not randomly distributed but is presented as a long-range self-similar correlation. By treating these data as a fractal dimension, it is now possible to determine changes in the long-term correlation of different behavioral sequences involved in nest maintenance owing to Pb exposure, both before and after adult males attain reproductive maturity. The authors hypothesized that the scaling exponent of this fluctuation varies in relation with environmental contaminants. Known Pb-induced changes in hormonal activity may account for changes in observed reproductive and nest maintenance behaviors. Pb-exposed fish exhibited higher levels of predictability in their behavioral sequences, i.e., they demonstrated an increase in the scaling parameter of the fluctuation {alpha}. However, if Pb was introduced after sexual maturity was observed, there was no significant difference in the scaling component {alpha}. Thus, the use of fractal dimension may provide a useful tool to analyze the effects of environmental contaminants and other stresses.
  • Article
    Full-text available
    When environment access is mediated through robotic sensors, field experience and naturalistic studies show robot handlers have difficulties comprehending remote environments - they experience what domain practi- tioners often call a 'soda straw'. This illustrates the keyhole effect in Human Robot Interaction, a CSE phe- nomena studied in the context of large virtual data space interfaces and the current research seeks to reduce this effect. A simulation for human-robot coordinated search and rescue was created based on WTC re- sponse experiences. Pilot studies showed traditional performance measures to be inadequate in analyzing control and exploration tasks therefore a novel analysis approach based on fractal path tortuosity was devel- oped. New interface concepts for helping remote observers perceive environmental affordances were then tested using the simulation environment and evaluation measures. These studies look to concepts based on Gibsonian principles to reduce keyhole effects in control interfaces to enhance remote functional presence in Human-Robot Coordination.
  • Article
    We investigated individual variation in juvenile bluegill sunfish's learning to forage on a novel food item. Ten bluegills were given the opportunity to forage on damselfly nymphs daily for 10 days. Each day three behavioral measures, latency until first capture, foraging success, and rate of prey capture, were recorded for each fish. By regression analysis each measure was fitted by an appropriate descriptive function, and the effects of time (i.e., days of exposure) and individual fish were tested. Time had a significant effect on all three measures, which indicated learning by the fish. Individual differences in learning ability were evident from a significant effect of fish on latency and capture rate. Because learning appears to be important in the development of foraging behavior in fish, we discuss the implications of individual variation in learning ability. (PsycINFO Database Record (c) 2012 APA, all rights reserved)
  • Article
    Quantitative evaluation of health impairments in wild chimpanzees was explored using fractal long-range correlations of behavioural sequences. The health status of 13 chimpanzees in the Mahale Mountains National Park, Tanzania was evaluated non-invasively using standard behavioural observation and parasitological analysis of stool samples. Based on these data, individuals were classified as being either healthy or sick. Behavioural sequences were analysed and shown to exhibit long-range power law correlations. The behavioural sequences of individuals in healthy and sick states were quantitatively evaluated using detrended fluctuation analysis of social and non-social behaviours within behavioural sequences. These values were compared and significant differences in long-range correlations were found between health states. Sex differences were also noted, with female social behaviour displaying a larger fractal dimension than male social behaviour. The fractal dimension of females declined significantly in sick individuals. This analysis of behaviour provides a new and effective non-invasive method to record and evaluate the general state of health and related stress of animals in the wild. Possible applications of this method in captive situations include the monitoring and evaluating of stress levels over time.
  • Article
    Most current methods for describing animal home range assume that it may be represented as a Euclidean type shape such as a bell shaped curve or a closed polygon. Landscape ecology has increasingly shown that ecological objects are more often highly fragmented and irregular. A fractal approach to description of animal home range was thus developed. For each point where the animal was observed, a circle centered on this point was first laid down to represent the area searched for prey by the animal during a short time interval. In this way the behavior of the animal and differences between species can be represented. Next, a fine grid is laid over the map and the height of each grid square computed by the number of circles that overlap that square. Then, the fractal dimension of the resultant 3-D surface is calculated at several scales. From an analysis of data from a hawk, the existence of perching behavior can be inferred, as well as the observation that at coarse scales the hawk behavior is self-similar and resembles a random walk. The home range thus analyzed in no way resembles a closed figure such as a polygon because it is highly fragmented. Further analysis showed that the fractal measures are relatively insensitive to sample size and to measurement error. Code is included for performing the analyses.
  • Chapter
    "...a blend of erudition (fascinating and sometimes obscure historical minutiae abound), popularization (mathematical rigor is relegated to appendices) and exposition (the reader need have little knowledge of the fields involved) ...and the illustrations include many superb examples of computer graphics that are works of art in their own right." Nature
  • Article
    This paper presents the background to and some initial results of an attempt to develop the basis for quantitative performance measures of robot behaviour. First, an example of a simple robot behaviour is used to motivate the need for a dynamical systems approach to the understanding and investigation of robot behaviour. The background and initial theoretical developments necessary for defining some appropriate quantitative measures are then presented. Finally, an example of the application of one of the techniques proposed is presented, together with a discussion of the practical difficulties involved and the future prospects of the presented approach.