Proceedings of the 2000 IEEE,RSJ

International Conference on
lntelligent Robots and Systems

A Comparative Study of Sensor-Based Path-Planning Algorithms

in an Unknown Maze

Hiroshi Noborio, Keiichi Fhjimura, Yohei Horiuchi

Department of Engineering Informatics
Osaka Electro-Communication University
Hatsu-Cho 18-8, Neyagawa, Osaka 572-8530, Japan
[email protected]

Abstract stacle has simple shape such as circle in an uncertain

cluttered environment, we proposed a near-optimal
I n general, an unknown maze has few collision-free sensor-based path-planning algorithm Simple and cal-
path to a destination. Therefore, i f a mbot supervised culated a small competitive ratio between lengths of
by the classic sensor-based path-planning algorithms the optimal path and a selected path as 1.58 [2].Oth-
Bug2, C l a s s l , A l g l , Alg2 repeatedly enters into long erwise, especially in an unknown huge maze, we have
local and global loops excluding and including a des- no proof to calculate a small competitive ratio and did
tination (goes out of its true way), respectively. For not find a near-optimal sensor-based path-planning al-
example, an Algl and Alg2, we can point out a case gorithm.
that a robot always enters into a global loop one time, Many algorithms have been aggressively proposed
and also in Bug(a1ter.) and Classl(alter.), we can for solving this problem. In the historical order, Lumel-
find another case that a robot frequently joins a local sky initially proposed Bug1 and Bug2 whose path
loop many times. A complicated maze usually includes lengths are denoted by D+1.5CiPi and D + C i T P i on
such cases, and therefore a robot arrives at a destina- the worst, respectively [3] ( D : the Euclidean distance
tion via a very long collision-free path. To overcome between initial and target points, Pi: the perimeter
this, we revisit an algorithm H D - I whose following of an i-th encountered obstacle, ni: the number of
direction is adequately changed by the trial and error. encountering an i-th obstacle). However, many peo-
I n H D - I , a robot hardly selects a n inadequate direc- ple guess that the latter algorithm selects a shorter
tion and consequently decreases a probability to enter collision-free path until a destination than the former
into global and local loops. algorithm on average. Therefore, Bug2 has been re-
peatedly modified. For example, Sankaranarayanan
modified Bug2 as Algl whose worst length is D +
2CiPi by eliminating a repetitive loop (i.e., a side
1 Introduction trip to a destination) [4]. On the other hand, one
of the authors proposed C l a s s l whose worst length is
In the past decade, many sensor-based path-planning
finitely bounded but is not evaluated, and also inves-
algorithms have been proposed [ l ] . In the sensor-
tigates convergence of a robot to its destination theo-
based path-planning, a robot always selects one of two
retically as the metric condition [5],[6]. Also, Sankara-
behaviors, that is, going straight t o its destination and
narayanan modified Classl as Alg2 whose worst length
avoiding an encountered obstacle. The behaviors are
switched if a robot hits or leaves an unknown obsta-
+
is D 2&Pi by eliminating a repetitive loop [7]. In
general, we cannot determine whether C l a s s l is supe-
cle. Here, as long as leave positions monotonically
rior to Bug2 or not, whether Algl is superior to Bug2
or asymptotically come close to a destination, con-
or not, and whether Alg2 is superior to C l a s s l or not
vergence of a robot to its destination is completely
on average. For example, a path selected by Bug2
ensured in an uncertain environment. This is called
and C l a s s l is frequently shorter than that selected
a metric condition. However, because of no environ-
by A l g l and Alg2, respectively, in the same unknown
ment information, a mobile robot supervised by al-
environment (see examples in [8]). Recently, one of
most all the classic algorithms unfortunately goes out
the authors proposed Bug2(alter.) and Classl(a1ter.)
of its way to a destination. Except for this, if each ob-

-909-
02000 IEEE.
0-7803-6348-5/00/$10.00

Authorized licensed use limited to: Cranfield University. Downloaded on April 30,2021 at 01:23:34 UTC from IEEE Xplore. Restrictions apply.
(their original algorithms are called Bug2(const.) and representing a complicated huge maze (Fig. 1). First
Classl(const.)) [8]. The difference is for a robot to se- of all, a 2-D uncertain environment is defined by a
lect alternatively clockwise and counter-clockwise di- digital map [zm,gm](m= 1,..,,2'). All the cells are
rections following an uncertain obstacle. By the al- built by dividing X and Y axes into an arbitrary r e s
ternative walk, a robot quickly arrives at a destina- olution. According t o the division, a digital map is
tion, whose property is theoretically ensured on av- built by 4' cells (nodes). Then, all cells (nodes) are
erage. However, if a robot selects a bad direction as classified into two categories such as Free and Obs
an initial direction following an uncertain obstacle, all cells (nodes). If and only if a cell (node) does not
the improved algorithms probably fail in seeking for intersect any obstacle, the cell (node) is Free, oth-
a shorter collision-free path until a destination. This erwise, the cell (node) is Obs. Secondly, a robot is
frequently occurs in a maze whose shape is complex. expressed by a cell (node), and is able to move one of
Independently, one of the authors proposed a three neighbor cells (nodes) in the digital map. In general,
or more dimensional sensor-based path-planning algo- eight cells (nodes) are prepared for neighbors. Two
rithm H D - I for a nonholonomic mobile robot and neighbor cells (nodes) are implicitly connected by an
a robotic manipulator [9],[10],[11]. To follow a three edge. Furthermore, the distance (cost) of an edge be-
or more dimensional obstacle, H D - I always learns tween horizontal and vertical neighbors is defined by
many directions by trial and error and then finds a one, and the distance (cost) of an edge between diag-
good direction for a robot to arrive at a destination.
The learning is useful for selecting a shorter collision-
onal neighbors is defined by a square route two
A total distance between start and goal cells (nodes
(a).
free path in a huge uncertain maze because it prevents ns and n9) is calculated by summing these neighbor
a robot from entering into a loop (i.e., a side trip to a distances.
destination).
The paper is organized as follows: Section 2 de-
scribes a robot and its unknown environment. In sec-
tion 3, we describe the seven sensor-based path-planning
algorithms BugZ(const.),Bug2(alter.),Class1 (const.),
Classl(alter.), A l g l , Alg2 and H D - I . In section 4,
we compare paths and lengths selected by these algo-
rithms in two kinds of 2-D uncertain mazes and con-
clude H D - I is the best algorithm to select a shorter
collision-free path until a destination on average. Fi-
nally, section 5 presents a few concluding remarks.

Figure 2 : (a) A collision-free path between n, and ng via many

leave and hit points (hj and I j ) selected in CIassl(const.). (b)
Another path between n. and ng via many leave and hit points (h,
and I j ) selected in Classl(altet.).

3 Seven Kinds of Algorithms

Figure 1 : A cell robot and its 2-D digital map with Free and O b s
cells (nodes). In this section, we briefly explain the sensor-based
path-planning algorithms BugS(const.), Bug2(alter.),
Classl(const.), Classl(a,lter.), A l g l , Alg2, H D - I
and their differences.
2 A Mobile Robot and Its Uncertain 3.1 The algorithms Classl(const.) and
Environment Classl(alter.)

In this paper, we adopt a cell decomposition ap- In this paragraph, we explain details of a classic
proach t o path planning, which has a flexibility for algorithm Classl(const.) and its revised algorithm

-910-

Authorized licensed use limited to: Cranfield University. Downloaded on April 30,2021 at 01:23:34 UTC from IEEE Xplore. Restrictions apply.
Classl(a1ter.) (Fig.2) [5],[6]. 3.2 The algorithms Bug2(const.) and
1. Set j = 1, and set a robot node np and a leave Bug2(alter.)
node l j - 1 and a closest node cj-1 by a start node n,.
2. Select a neighbor node ni of np,which the seg- In this paragraph, we briefly explain the difference
ment between centers of np and ng intersects in a dig- between Class1 and Bug2. The difference is the ex-
ital map. istence of the segment condition such that a robot ni
(2a) If ni is a goal node n g ,exit with success. cannot leave an obstacle if it is not t o be on the seg-
(2b) If ni is Free, set np by ni and then be con- ment between centers of n, and ng (Fig.3). The details
tinued, otherwise, set a hit node hj and a closest node of Bug2(const) and Bug2(alter.) are addressed in [3]
+
c j ( j = j 1 ) by np and go to step 3. and [8], respectively.
3. Select a neighbor Free node ni of np by one of (3b) If ni is to be on the segment between
the clockwise and counter-clockwise orders based on centers of np and ng and also ni is closer t o ng than
the chain coding, which is adjacent t o a neighbor Obs cj , a robot checks if the direction from ni to n9 is Free
node until one of the following occurs. The following (that is, a few nodes from ni to ng are Free). If it is
order is fixed in advance in Classl(const.), on right, regard ni as a leave node lj and return to step
the other hand, it is alternatively changed in 2, otherwise, reset cj by ni.
Cla,ssl (alter.) (Fig.2).

: I .

t A h i t point
o A IMVO point

-/ '-' - 1st path

2nd path
Figure 4 : A collision-free path betwccn n, and ngvia many leavc
(a) (b) and hit points ( h , and l j ) sclected in A l g l .

Figure 3 : (a) A collision-frcc path betwccn n. and ng via many

lcavc and hit points ( h , and 1,) selcctcd in Bug2(const.). (b) An-
other path bctwccn ns and ng via many lcavc and hit points ( h j 3.3 The algorithm Algl
and 1,) sclcctcd in BugZ(alter.).
In this paragraph, we briefly explain an algorithm
Algl (its detail is addressed in [4])(Fig.4). The al-
(3a) If n, is a goal node n g ,exit with success. gorithm Algl is designed under Bug2(const.) in order
(3b) If n, is closer to n9 than the closest node c,, to get a shorter path until a destination frequently.
reset c, by n,. Then, if the direction from n, to ng is Usually if a robot goes straight to a destination or fol-
Free (that is, a few nodes from n, to n9 are F r e e ) , lows an encountered obstacle one time, Algl equals to
regard n, as a leave node I, and return to step 2. Bug2(const.). Otherwise, that is, if a robot meets a
(3c) If n, equals t o the last hit node h,, exit with previous leave or hit point, the robot returns to the
failure. In both algorithms Classl(const.) and last hit point via a shorter route, and then reverses
Classl(alter.), leave and hit nodes ( I , and h,) mono- a direction following an obstacle. Then, after leaving
tonically come close to ng,and consequently nparrives the obstacle, Algl equals to BugZ(const.). As an ex-
at ng finally if a path (a sequence of Free nodes) be- ample, when a mobile robot joins a cycle, e.g., a cycle
tween n, and ng exists in a given map. Otherwise, from a to a via b, the robot backs t o the last hit point
a point robot np automatically recognizes no solution h3(= b) by selecting the shorter route of clockwise and
(path) between n, and n9 in an unknown map by re- counter-clockwise ones (the counter-clockwise route is
turning the last hit node h, without finding a leave selected in Fig.4). Then, after backing to the last hit
point 1,. point h3(= b ) , the robot follows an unknown obstacle

-911-

Authorized licensed use limited to: Cranfield University. Downloaded on April 30,2021 at 01:23:34 UTC from IEEE Xplore. Restrictions apply.
in the opposite tracing order. Then after leaving the 3.5 The algorithm HD-I
obstacle, Algl equals to BugZ(const.).
In this paragraph, we focus on B F and Classl
which do not consider a past cost g ( n p ) from a start
node n, t o a present node np. Then, we design a learn-
ing sensor-based path-planning algorithm H D - I by
mixing modified B F and Classl (Fig.6) [9]. As men-
tioned previously, the sensor-based pat h-planning con-
sists of two procedures. One procedure is for a robot
to go straight to a goal. Another procedure is for a
robot t o avoid an uncertain obstacle. In the algorithm
H D - I , the former procedure and the switching are
designed by the modified Classl, and the latter pro-
cedure is designed by the modified B F .
( 1 ) H D - I uses B F in order for np to trace an en-
countered obstacle by clockwise and counter-clockwise
orders. That is, if a robot node np meets a hit node hj
around an uncertain set of Obs nodes, B F starts (be-
Figure 6 : A collision-free path between n. and no via many leave
and hit points ( h j and l j ) selected in Alg2.
gins to input a found or expanded node into O P E N
or C L O S E D list), and if np reaches a leave node l j
around it, B F ends (cleans O P E N and C L O S E D
lists).
3.4 The algorithm Alg2 (2) H D - I uses B F in order for npto trace bound-
ary of an unknown obstacle, i.e., an uncertain set of
Obs nodes. First of all, in a 2-D digital map, the
In this paragraph, we briefly explain an algorithm number of nodes neighboring np is denoted by 32 - 1.
Alg2 (its detail is addressed in [7]) (Fig.5). The algo- Secondly, the tracing starts if a robot node np finds a
rithm Alg2 is designed under CZassl(const.) in order Obs node as its neighbor ni on the segment between
np and ng. Therefore, np is Free and ni is Obs. In
to get a shorter path until a destination frequently. this case, there is at least one Free node neighbor-
The idea to shorten is just the same with Algl. For ex- ing np,which neighbors the set of Obs nodes. All the
ample, when a mobile robot joins a cycle, e.g., a cycle Free nodes are inserted into O P E N .
from a to a via b, the robot backs t o the last hit point (3) To select a hopeful node n* from O P E N , we
h5(= b ) by selecting the shorter route of clockwise and evaluate each node n* by summing a moving cost
counter-clockwise ones (the counter-clockwise route is m(n*)from np to n* and a future cost h(n*)from n*
to ng. The future cost is straightforwardly calculated
selected in Fig.5). Then, after backing t o the last hit by the piecewasel2 metric along the segment between
point h5(= b ) , the robot follows an unknown obstacle n* and ng. On the other hand, the moving cost is
in the opposite tracing order. Then, after leaving the defined by multiplying a and the length of a truly
obstacle, Alg2 equals to Classl(const.). traversed route between np and n*.
1. Set j = 1, and set a robot node np and a leave
node lj-1 and a closest node cj-1 by a start node n,.
2. Select a neighbor node ni of a robot node np,
which the segment between np and n9 intersects in a
digital map.
(2a) If ni is a goal node n9,exit with success.
(2b) If ni is Free, set np by ni and then be con-
tinued, otherwise, set a hit node hj and a closest node
cj(j = j + 1 ) by np and go to step 3.
3. Put the node hj on a list called O P E N of un-
expanded nodes.
4. If O P E N is empty, exit with failure; no path
ex$?.
v
,
t
-2ndpath
Remove from O P E N a node np at which q is
minimum (break ties arbitrary), and place it on a list
called C L O S E D t o be used for expanded nodes.
Figure 6 : A collision-free path between ns and no via many leave 6. Select each neighbor Free node ni of np,which
and hit points (hj and I j ) selected in H D - I . neighbors on the set of Obs nodes. For every neighbor
ni :
(sa) If ni is a goal node ng,exit successfully.

-912-

Authorized licensed use limited to: Cranfield University. Downloaded on April 30,2021 at 01:23:34 UTC from IEEE Xplore. Restrictions apply.
'Jhble 1 : I'ath lengths from four start positions n, to a goal position n g in Classl, Bug& Algl, Alg2, and HD - I in a 2-D uncert.ain maze.

(6b) If ni is closer to ng than the closest node cj,

reset cj by ni, and if a few nodes from n, to ng are
Free, regard ni as a leave node l j . Then, clear O P E N
and C L O S E D and return to step 2.
( 6 c ) If ni was neither on O P E N nor on C L O S E D ,
register it t o O P E N and generate ni with the pointers
back to np.
(6d) Evaluate each node n* in O P E N as follows:
Firstly, calculate a future cost h(n*)by the piecewiseL2
metric along the segment between n* and ng, Sec-
ondly, calculate a moving cost m(n*)by the piecewasel2
metric along a route already traveled between np and
n*. Finally, assign q(n*)= axm(n*) h(n*)(O< +
Q < 1) to each node n* in O P E N , and then return to
step 4.
In the algorithm H D - I , leave and hit nodes ( l j
and h j ) monotonically converge to a goal node (n,)
Figure 8 : (a),(b) A mobile robot joins a global loop one time and
in a 2-D digital map. In addition, np always finds consequently does not arrive a t a destination quickly in Algl and
an adequate leave node l j around an encountered set Alg2, respectively.
of Obs nodes as long as a sequence of Free nodes
from ns to ng exists in a given digital map. For this
reason, if such a sequence exists, npcertainly arrives at
ng,otherwise, np sees no path since O P E N becomes
empty.

Figure 7 : (a),(b) A mobile robot joins similar global loops several

times and extremely increases a driving distance to a destination in Figure 0 : (a),@) A mobile robot Joins a loop several
BugZ(const.) and Classl(const.), respectively. times and consequently its collision-free path becomes very long
in BugZ(a1ter.) and Classl(alter.),respectively.

-913-

Authorized licensed use limited to: Cranfield University. Downloaded on April 30,2021 at 01:23:34 UTC from IEEE Xplore. Restrictions apply.
 Figure 12 : (a) A collision-free path from n,:(7,7) to ng:(277,255) selected by Classl(alter.)[clockcwise]in a 2-D uncertain maze. (b) A
path between the same pair selected by Algl[eloekwise] in the maze. (c) A path between the same pair selected by HD - I in the maze.

Figure 13 : (a) A collision-frce path from n.:(506,506) to ng:(277,255) selected by BugZ(alter.)[c -clockwise] in a 2-D unknown maze. (b)
A path between the same pair selected by Algl[c - clockwise] in the maze. (c) A path betwcen the same pair selected by HD - I in thc
maze.

4 Discussions
In an unknown maze, a robot enters into global
and local loops several times and consequently gen-
erates a very long path until a destination. For ex-
ample, the most classic algorithms Bug2(const.) and
Classl(const.)pick up very long collision-free paths
in an uncertain environment (Fig.7). A few of classic
sensor-based path-planning algorithms partially over-
come this problem as follows: In the algorithms Algl
and Alg2, a robot returns t o the closest position to
Figure 10 : A mobile robot does not easily join a global loop and a destination and reverses its direction following an
consequently arrives at a destination quickly in H D - I . obstacle (i.e., escapes from a global or local loop) if
the robot encounters a position visited already (i.e.,
meets the loop). By the procedure, a robot never en-
\"==la
ters into the same loop after this. In the algorithms
Bug2(alter.)and Classl(alter.),a robot alternatively
changes a direction following an encountered obstacle.
This theoretically brings decreasing a probability that
a robot joins a global loop. Moreover, in Single, we
succeeded in mixing advantages of Algl or Alg2 and
Bug2(alter.) or CZassl(a1ter.)in an uncertain maze
with an unique collision-free path to a destination [12].
However, because of a contradiction, we did not
succeed in mixing advantages of Algl or Alg2 and
I3
BugZ(a1ter.) or Classl(a1ter.) in an uncertain maze
with two or more collision-free paths. In general, a
weak point of Algl or Alg2 is for a robot to enter into
Figure 11 : A mobile robot does not easily join a local loop and a global loop one time (Fig.8), on the other hand, a
Consequently arrives at a destination quickly in HD - I . weak point of Bug2(alter.) or Classl(a1ter.) is for a

-914 -

Authorized licensed use limited to: Cranfield University. Downloaded on April 30,2021 at 01:23:34 UTC from IEEE Xplore. Restrictions apply.
 Table 2 : Path Icngths from four n8 to one ng in C l a s s l , Bug2, A lgl, Alg2, and H D - I in another 2-D unccrtain mazc.

robot to enter into a local loop several times (Fig.9). lowing an encountered obstacle. However, H D - I
As contrasted with this, we proposed a sensor-based always learns an adequate direction by the trial and
path-planning algorithm H D - I whose trial and er- error, and therefore selects a shorter collision-free path
ror learns a hopeful direction following an encountered until a destination, which has no connection with an
obstacle. In order to find a leave position around an ini ‘a1di ction
unknown object, the algorithm H D - I uses a mod- 8econ8?y, another huge uncertain maze is built at
ified version of the famous Best - First algorithm. random and then three different collision-free paths
The reason why we adopt B F is the similarity to the are picked up from ns:(9,502) t o ng:(275,263) by Classl
sensor-based path-planning algorithm with the met- (alter.) [clockwise],Alg2 [clockwise]and H D - I in an-
ric condition. That is, both do not consider any cost other unknown maze illustrated a t Fig.l4(a),(b) and
from a start position t o a present position. In other (c), respectively. In addition, three different collision-
words, both consider only a future cost from a present free paths are selected from ns:(502,9) t o n,:(275,263)
position to a target position, which is called heuris- by Bug2(const.) [c-clockwise], Algl [c-clockwise] and
tics. The modified algorithm B F always selects a H D - I in the same maze shown at Fig.lS(a),(b) and
most hopeful point around an encountered obstacle.
This trial and error flexibly changes a direction to (c), respectively.. Furthermore, by changing n,, we
find an adequate leave point around an obstacle. This obtain the other results (Table 2, long means path
probably overcomes the previous weak points because length is over 30000). In the rest three results, our al-
the trial and error does not allow a robot to go away gorithm H D - I always learns an adequate direction
from a destination (Fig.10 and 11). by the trial and error, and therefore selects a shorter
collision-free path until a destination.
As a result, on the average, our learning sensor-
based path-planning algorithm H D - I catches the
5 Simulation Results shortest or a shorter path in a huge uncertain maze
whose shape is complicated. Finally, all programs
In this section, we compare paths and lengths se- are running in a personal computer Gateway GP6-
lected by typical sensor-based path-planning algorithms 400 (Windows 98 with Open GL, Pentium I1 400MHz,
Bug2, Classl, Algl, Alg2, and H D - I in two kinds 256MB main memory).
of uncertain huge mazes and then select the best algo-
rithm. A 2-D digital map [x,y] is defined as Fig.1, and
many obstacles whose shape is complex are colored by
gray, and a selected path is colored by black. 6 Conclusions
Firstly, a huge uncertain maze is randomly con-
structed, and then three different collision-free paths In this paper, we compared typical seven sensor-
are selected from ns:(7,7) to ng:(277,255) by Classl based path-planning algorithms in two kinds of uncer-
(alter.) [clockwise], Algl [clockwise] and H D - I in tain mazes, and regarded the algorithm H D - I as the
an uncertain maze described at Fig.l2(a),(b) and (c), best on the average. This was ascertained by theoret-
respectively. In addition, three different collision-free ical discussions and simulation results. In future, we
paths are selected from ns:(506,506) to ng:(277,255) will be seeking for the algorithm whose superiority is
by Bug2(alter.) [c-clockwise], Algl [c-clockwise] and theoretically ensured.
H D - I in the same maze shown at Fig.l3(a),(b) and
(c), respectively. Furthermore, if we change n,, we
obtain the other results (Table 1, long means path References
length is over 30000). In the results, our algorithm
H D - I selects the shortest or shorter path. Especially, [I] H.Noborio, ”On a sensor-based navigation for a mobile
path lengths of all the previous algorithms except for robot,” Journal of Robotics Mechatronics, vo1.8, No.1,
H D - I extremely depend on an initial direction fol- pp.2-14, 1996.

-915-

Authorized licensed use limited to: Cranfield University. Downloaded on April 30,2021 at 01:23:34 UTC from IEEE Xplore. Restrictions apply.
Figure 14 : (a) A eollision-free path from n.:(9,502) t o n,:(275,263) selected by Classl(alter.)[clockwise]
in another 2-D unknown maze.
(b) A path bctween the same pair selected by AlgP[elockwise] in the maze. (c) A path between the same pair selcctcd by H D - I in the
maze.

Figure 15 : (a) A collision-free path from n.:(502,9)to n,:(275,263) selected by Bug2(const.)[c- clockwise] in another 2-D unknown maze.
(b) A path between the same pair selected by Algl[c - clockwise] in the maze. (e) A path between the same pair selected by H D - I in the
maze.

[2] Hiroshi Noborio and Kenji Urakawa, “ A near-optimal [8] H.Noborio, T.Yoshioka, and S.Tominaga, ” O n the sensor-
sensor-based navigation among 2-D uncertain obstacles based navigation by changing a direction t o follow a n en-
with simple shape,” Proc. of the 1999 IEEE Int. Conf. on countered obstacle,” Proc. of the IEEE/RSJ Int. Conf.
Robotacs and Automatton, pp.355-360, April 1999. on Intelligent Robots and Systems, pp.510-517, September
1997.
[3] V.J.Lumelsky and A AStepanov, ”Path-planning strate-
gies for a point mobile automaton moving amidst unknown (91 H.Noborio, Y.Maeda, and K.Urakawa, ”Three or more
obstacles of arbitrary shape,” Algonthmcca, V01.2, pp.403- dimensional sensor-based path-planning algorithm HD-I,”
430, 1987. Proc. of the IEEE/RSJ Int. Conf. on Intelligent Robots
and Systems, pp.1699-1706, October 1999.
[4] ASankaranarayanan and M.Vidyasager, ” A new path plan-
ning algorithm for moving a point object amidst unknown
[lo] H.Noborio, LYamamoto, T.Komaki, ”Sensor-based path-
planning algorithms for a nonholonomic mobile robot”,
obstacles in a plane, Pmc. of the 1990 IEEE Int. Conf. on
Proc. of the IEEE/RSJ Int. Conf. on Intelligent Robots
Robotzcs and Automatton, pp.1930-1936, May 1990.
and Systems, October 2000 (to appear).
[5] H.Noborio,”A path-planning algorithm for generation of an [ll] J . Pearl, ”Heuristics”, Addision Wesley, October 1985.
intuitively reasonable path in a n uncertain 2D workspace”,
Proc. of the Japan-USA Symposaum on Flextble Automa- [12] H.Noborio, M.Kadowaki, and K.Urakawa, ” A near-optimal
tzon, pp.477-480, July 1990. sensor-based motion-planning algorithm for parts mating,”
Proc. of the IEEE/RSJ Int. Conf. on Intelligent Robots
[6] H.Noborio, ” A sufficient condition for designing a family and Systems, pp.510-517, October 1998.
of sensor-based deadlock-free path-planning algorithms,”
Journal of Advanced Robotacs, Vo1.7, No.5, pp.413-433, [13] S S u n d a r and ZShiller, ”Optimal obstacle avoidance based
January 1993. on the hamilton-jacobi-bellma? equation,” IEEE h m .on
Robotics and Automation, Vol. 13, No.2, pp.305-310, 1997.
[7] ASankaranarayanan and M.Vidyasagar, ” P a t h planning
(141 1.Kamon and E.Rivlin, ”Sensory-based motion planning
for moving a point object amidst unknown obstacles in a
with global prooES,’’ IEEE l h m . on Robotics and Automa-
plane: a new algorithm and a general theory for algorithm tion, Vo1.13, No.6, pp.814822, 1997.
development,” Proc. of the 29th Conf. on Decision and
Control, pp.1111-1119, December, 1990.

-916-

Authorized licensed use limited to: Cranfield University. Downloaded on April 30,2021 at 01:23:34 UTC from IEEE Xplore. Restrictions apply.