474

advertisement
Available online at www.sciencedirect.com
ScienceDirect
Procedia Engineering00 (2014) 000–000
www.elsevier.com/locate/procedia
“APISAT2014”, 2014 Asia-Pacific International Symposium on Aerospace Technology,
APISAT2014
Research progress and future development on mission planning
technologies of planetary rover
Wang Qionga,b,*, Yu Dengyunc, Jia Yangb
a
Lunar Exploration and Space Engineering Center,No.12 Chegongzhuang Street, Xicheng District, Beijing 100037, China
b
Beijing Institute of Spacecraft System Engineering, No. 5142 Mailbox, Beijing 100094, China
c
China Aerospace Science and Technology Corporation, No.16 Fucheng Road, Haidian District, Beijing 100048, China
Abstract
Roving exploration becomes a major method of further exploration to target planet after preliminary circling exploration. Mission
planning is to search feasible action sequence and trajectory of the rover from initial state to target state, satisfying complex
constraints such as terrain, energy, thermal control and communications. Hence it is the basis and key of controlling rover to
explore the planet safely and effectively. Global path planning and local path planning could be seen as a special case of mission
planning. In this paper existing planetary rovers, prototypes and related navigation technologies are briefly summarized, followed
by an introduction of various mission planning algorithms. Finally several areas for future research are specified.
© 2014 The Authors. Published by Elsevier Ltd.
Peer-review under responsibility of Chinese Society of Aeronautics and Astronautics (CSAA).
Keywords:Exploration robot; Mobile robot; Mission planning; Path planning
1. Introduction
Planetary rover is a kind of spacecraft which could adapt planetary surface environment, move on the planetary
surface, and carry out exploration, sampling, transport and such complex missions [1]. For short hereafter it is called
rover. It has two kinds of function: (1) to help human accomplish tasks that are too difficult or uneconomical for
human, (2) to improve precision, expand range and increase means of scientific exploration.
* Corresponding author. Tel.: +86-10-88306151; fax: +86-10-88306170.
E-mail address: wangq2006@163.com
1877-7058 © 2014 The Authors. Published by Elsevier Ltd.
Peer-review under responsibility of Chinese Society of Aeronautics and Astronautics (CSAA).
2
Wang Qiong/ Procedia Engineering00 (2014) 000–000
By roving exploration much deeper and more precise results could be obtained than by orbital exploration, and
much larger range could be reached than by fixed landing exploration. These results could complement and prove
each other. Hence, roving exploration becomes a major method of further exploration to target planet after
preliminary circling exploration, and is taken seriously by many countries. Various prototypes were developed and
several rovers were successfully employed in planet exploration missions, including three manned lunar vehicles,
three robotic lunar rovers and four robotic mars rovers. Moreover, ESA ExoMars mars rover, Japanese SELENE-2
lunar rover, Indian Chandrayaan-2 lunar rover and several Google X-Prize participant rovers are under developing.
These programs promoted the development of mission planning technology.
In this paper, existing planetary rovers, prototypes and related navigation technologies are overviewed, followed
by an introduction of various mission planning algorithms. Finally several areas for future research are specified.
2. Outline of planetary rovers
2.1. Robotic rovers that successfully carried out planet exploration missions
Lunokhod-1 and Lunokhod-2 were developed by the Soviet Union. Lunokhod-1 landed on the moon on
November 1971, becoming the first lunar rover in human history. Lunokhod-2 landed on January 1973. Their
objectives were gathering images of lunar surface, carrying out soil physical test and chemical analysis in 90-day
lifetime. In fact, Lunokhod-1 had worked for 322 days and Lunokhod-2 for 125 days.
Sojourner, Spirit (MER-A), Opportunity (MER-B), Curiosity came from the USA. Sojourner landed on mars on
July 1997 and became the first mars rover of human. It was designed to travel 10m in 7-day lifetime, but finally it
traveled 104m around the lander in 83-sol life, during which wheel-soil interaction was determined and autonomous
navigation and hazard-avoiding technologies were tested. The lessons learned were used for improving future rover
designs. Spirit and Opportunity landed on mars both on January 2004. Their designed lives are both 90 sols, but
actually Spirit had worked for more than seven years and Opportunity is still working. They had found the evidence
of past water activities on the mars. Curiosity is the most powerful planetary rover so far. It landed on the Gale
Crater region of mars on August 2012. Its designed life is two Martian year and now has completed more than half
after finding that Mars once had environmental conditions favorable for microbial life.
Chinese rover Yutu landed on Mare Imbrium of lunar on December 2013. The scientific objectives of Yutu
mainly include lunar surface topography and geological survey, lunar surface material composition and resource
survey. Its designed life is three months, but it encountered operational difficulties in the second lunar day and still
gathered some useful data in the following several months.
2.2. Robotic rover prototypes
Through the development of various robotic rover prototypes, experiences, lessons and technical reserves were
accumulated for formal missions.
Rocky series have superior mobility performance because of their six-wheeled Rocker-bogie suspension system.
Among them, Rocky 7 and FIDO are the most mature. Rocky 7 [2] improved Sojourner to test new technologies for
long distance traverse. FIDO [3] tested a series of new technologies for MER missions including navigation and
control, ground planning, command and visualization tools. Marsokhod [4] comprises three parts, which could
relatively rotate or incline. Hence it could passively adapt rough terrains. Marsokhod was used for test platform by
several institutes. RATLER [5] was a test platform of lunar rover tele-operation technology. It consists of two parts
that could relatively rotate, and could keep four wheels in touch with the ground. Nomad [6] was developed to test
the capability of long-distance, long-duration exploration. A transforming chassis along with an averaging
suspension system and self-contained wheel modules were adopted by Nomad. Hyperion and Zoe [7] were test
platforms of sun-synchronous lunar polar exploration. In order to keep the rover always under the sun, its paths
could be planned according to current rover position, current time and solar azimuth. A tracked mobility system was
adopted by Nanokhod [8], which obtained energy and swapped data from/with the lander through umbilical cables.
SHRIMP [9] adopted a unique chassis layout including a front wheel, a rear wheel and four middle wheels.
Wang Qiong/ Procedia Engineering 00 (2014) 000–000
3
In China, various robotic lunar rover prototypes were developed by some industrial departments, institutes and
universities. Among them, a six-wheeled, Rocker-bogie lunar rover prototype developed by China Academy of
Space Technology was relatively the most mature. It was deployed in the desert to perform several demonstrations
and experiments of new technologies, which was finally applied on Yutu rover.
2.3. Manned planetary vehicles
Lunar Roving Vehicle (LRV) was a four-wheeled, two-seater lunar vehicle used for transporting astronauts and
equipments on lunar surface. LRV was used in Apollo 15 mission and landed on the moon on July 1971, as the first
manned lunar vehicle in human history. Subsequently LRV was used in Apollo 16, Apollo 17 missions. Space
Exploration Vehicle (SEV) [10] is a new generation of modularized, multi-mission space vehicle, which could be
used for in-space missions as well as surface exploration on planetary bodies. The SEV surface concept has a small
pressurized cabin mounted on a twelve-wheeled chassis. ATHLETE [11] was a six-wheel-legged all-terrain mobility
platform, whose main function was transporting large modules such as habitation modules. RAMA [12] was a
pressurized four-wheeled planetary vehicle designed by ESA. It could work independently or combine other
vehicles as a base. Moreover, MOBEV series, LOTRAN, Rover First, Daylight Rover, LPMR, MSTS and other
concepts and prototypes of manned planetary vehicle were proposed or developed.
3. Outline of navigation technologies on planetary rovers
Rover navigation is that rover senses environment and its own status by sensors, and realizes autonomous move
to target in environment with obstacles. It mainly focuses on three aspects [13]: (1) to obtain the rover position,
direction and environment information by some detection methods, i.e., environment sensing; (2) to process those
obtained information and build environment model by some algorithms, i.e., environment modeling; (3) to find an
optimum or approximate optimum free path according to some criterion, i.e., path planning.
In environment sensing aspect, stereo vision based sensing modes were mostly adopted. Particularly, considering
the platform safety, close-range hazard avoidance sensors were deployed. Laser sensors were adopted in the past,
such as laser stripe sensors on Sojourner, laser rangefinder on Ratler. But on the latest rovers such as Opportunity,
Spirit, Curiosity and Yutu, hazard-avoidance cameras based on binocular stereovision became essential.
Correspondingly, in environment modelling aspect, three dimensional scene reconstruction methods based on
stereovision were mostly adopted. Environment images obtained by stereo cameras were used to generate navigation
map, after a series of processing steps including image correction, feature extraction, stereo matching, 3-D
information estimation. DEM was the most common navigation map and other types of map also existed. For
example, a traversability map indicating dangerous zones and safe zones was generated based on DEM by FIDO [3].
A more complicated goodness map was produced by Nomad to indicate how good it is to occupy a cell [6].
In path planning aspect, due to the onboard resource limitation, algorithm practicality depends on whether a good
balance among complexity, real-time performance and optimality could be reached. Early rovers only had simple
intelligence, and most works should be done through tele-operation by ground staffs with very low efficiency. For
instance, Lunokhod was completely navigated and controlled by ground operators; Sojourner could only perform
simple local reactive hazard-avoidance, and its waypoints were chosen by ground operators. As the emergence of
stronger sensors, onboard computers and more advanced GNC algorithms, rovers gradually had an improved
capability of autonomous navigation and control, as well as less manual interventions and higher efficiency. For
example, onboard global path planning algorithm Field D* and local path planning algorithm GESTALT were
adopted on Opportunity, Spirit and Curiosity [14]. Ground operators needed only assign exploration targets and
tasks of the next day. Rovers could plan paths, navigate themselves and move to targets autonomously.
4. Research progress of mission planning algorithms
Mission planning problem is to find a behavior sequence and a travel path from initial state to target state, given
initial conditions (including initial rover state, time and environment conditions), behavior constraints set and target
4
Wang Qiong/ Procedia Engineering00 (2014) 000–000
set (including target rover state, time, etc.) [15]. Rover state includes position, energy level, temperature level,
communications conditions and hold of other resources. The transitions between states are realized by behaviors,
including sensing, moving, charging, scientific exploration, data transmission, hibernation, and so on. Path planning
could be seen as a special case of mission planning if we only consider position for rover states, moving for
behaviors, terrain obstacles for constraints. According to the acquaintance of environment information, path
planning comprises global path planning based on complete prior information and local path planning based on realtime sensing environment information.
In practical use, mission planning usually deals with cases of low resolution, more than hundred meters’ distance
and dozens of hours’ duration. Its outputs are strategic waypoints and behavior sequence, which could provide initial
point, target point and basic planning direction for global path planning and local path planning within certain range.
Global path planning focuses on path searching, and deals with cases of medium resolution, dozens of meters’
distance and several hours’ duration. Local path planning focuses on obstacle avoiding, and deals with cases of high
resolution, several meters’ distance and dozens of minutes’ duration.
Because planetary rovers commonly have a wheeled mobility system with few degrees of freedom, its
configuration space has small dimensions. Moreover, planetary rovers work in the complex unstructured
environment of planetary surface, having limited resources but high safety demands, therefore deterministic
algorithms are usually adopted to achieve completeness even optimality and certain bounds on the quality of path
solution. In this paper, deterministic algorithms are mainly introduced.
4.1. Mission planning algorithms
RCW [16] is the mission planner of Sojourner. In RCW, path planning and behavior planning are almost
independent. Operators used RCW to estimate safe paths and select waypoints at intervals of 1-2 meters as
intermediate navigation goals, based on terrain traversability and geometric constraint of line-of-sight between
Sojourner and the lander. Behavior planning is performed by simple constraint checks, and interaction effects
between various behaviors are not considered.
MAPGEN [17] is the mission planner for MER missions. In early stage of the missions, operators used orbital
imagery and rover stereo imagery to manually select waypoints and estimate the optimal path according to terrain
traversability and sun, solar array and terrain geometry data. After adding Field D* algorithm into onboard software,
operators only need assign exploration targets and tasks of the next day, and the rover would plan path and generate
waypoints. MAPGEN adopts Simple Temporal Network (STN) constraint reasoning and scheduling technology to
plan behaviors. During arranging the traverse behavior, distance between waypoints is used to estimate duration,
energy distribution and other constraints. Hence, traverse behavior is loosely coupled with other behaviors, and
interaction effects between behaviors get more consideration. But MAPGEN also determines path first, then
arranges behaviors along the path. The effects of behaviors on path selection are not taken into account, so path
planning and behavior planning are still independent.
CLEaR [18] is a planning and execution system of position-distributed science measurements. Goal selection was
performed by operators, and paths between goals were given by the path planner. But the paths are only used to
estimate traverse distances and calculate corresponding expect time and energy costs, as constraints for behavior
planning. Through the field tests of Rocky 7, Rocky 8 and FIDO, this path planning algorithm was found to be too
simple. It was ignorant of time, resources limitations and geometric constraints, hence the eventual paths followed
by the rovers differ greatly to the generated paths, and the estimated time and resource costs were also inaccurate. In
CLEaR, traverse and other behaviors were closely interleaved, and time and resource allocations of each behavior
strongly affected the feasibility of others.
Derived from MAPGEN, ENSEMBLE [19] is the mission planner of Curiosity rover. It unifies the disparate
tools in MER, while providing a better user interface. Its core planner EUROPA2, the next generation version of
EUROPA used in MAPGEN, supports modeling of state and other resources.
These above-mentioned planners can be classified as mission planning methods based on resource scheduling,
focusing on behavior planning to solve resource conflicts. Path planning is only used for resource usage estimation
and plays an auxiliary role. The idea that independently plan paths and behaviors is not the same as practical
Wang Qiong/ Procedia Engineering 00 (2014) 000–000
5
situation of rover operation on planetary surface. Therefore, the mission planning results are less instructive for
exploration activities due to large estimation error.
Another research direction is path-based mission planning, which treats mission planning as a class of
generalized path planning problems. The main idea is generalizing the path planning algorithms from lowdimensional state space to high-dimensional state space, and performing constraint checks while searching paths. In
these “Traverse—Judge—Insert behavior—Traverse” reactive planning algorithms, behaviors are inserted along the
generated path according to trigger conditions. ISE [7], core of the TEMPEST planner, generalized D* algorithm to
high-dimensional state space and conducted backward mission-level path planning. Peng Song et al. [20] improved
and generalized A* algorithm, and proposed Real-time Greedy algorithm. Pei S et al. [21] applied SD* Lite
algorithm into mission planning, whose highlight is generating path in any direction. Xie Yuan et al. [22] utilized
the predictable global dynamic characteristic of lunar surface environment, and proposed a Predictable Dynamic
Environment algorithm and introduced a flexible plan concept. This algorithm was applied in Yutu mission. Song
Pei [23] proposed a mission planning method based on predicate logic, which was also used in Yutu mission. These
algorithms plan path and behaviors at the same time, and take into consideration of the behaviors’ effects on path
selection. They do well on path optimality, but in return, the results of behavior planning are only feasible solutions
other than optimal solutions.
4.2. Global path planning algorithms
Global path planning algorithms are mostly based on geometrical model search. The geometrical model of rover
environment space is built at first, and then some graph search algorithm is applied to obtain the optimal path.
Geometrical modeling methods mainly include roadmap methods and cell decomposition methods. Graph search
algorithms include Breadth-First, Depth-First, Iterative Deepening, Dijkstra, A* and other improved algorithms.
4.2.1. Roadmap methods
Visibility graph method [24] builds a visibility map that guarantees two connected points visible to each other,
and searches the shortest path on this map. The method is simple and intuitional, but has a complex process of graph
generation as well as low search efficiency. Tangent graph method [25] deletes lots of useless connection lines from
visibility graph, hence greatly reduces the search space. The method could find out the shortest path, but resulting
path is close to the obstacles leading to poor safety. Voronoi diagram method [26] limits the rover to keeping the
same distance with obstacles. This reduces the dimensions of configuration space and is conducive to searching
feasible path. Compared with Tangent graph method, Voronoi diagram method generates longer path, but keeps the
rover far away from obstacles. MAKLINK graph method [27] deletes unnecessary connections from visibility graph,
and constructs lots of largest convex polygons to generate connected graph. The method is flexible because that
change of initial point and target point would not reconstruct the graph, but construction of convex polygons is not
unique, hence the resulting path is indeterminate and non-optimal.
4.2.2. Cell decomposition methods
Cell decomposition methods decompose environment space into connected and non-overlapping cells, and search
an optimal free path on the adjacency graph constructed by these cells [28]. These methods could be divided into
two classes: exact cell decomposition and approximate cell decomposition. The former is complete but has a high
modeling cost. The latter is easy to construct, save and use, hence it is widely applied. But it has low storage
efficiency especially in environments containing large free space or obstacles. Moreover, due to the eightneighborhood limitation of rover heading, generated path is not optimal [15]. Aiming at above disadvantages of grid
method, various multi-resolution grid methods were proposed, such as framed-quadtree [29], weighted framedquadtree and weighted framed-octree [30], variable resolution grid [31], compact mesh representation [32].
4.2.3. Graph search algorithms
Dijkstra algorithm [33] is a classical shortest-path search algorithm, which introduces evaluation function based
on Breadth-First algorithm. A* algorithm [34] adds heuristic function to Dijkstra. A* is best-first, efficient and easy
6
Wang Qiong/ Procedia Engineering00 (2014) 000–000
to implement and guarantees to find out the best solution, hence it is widely used. But local changes of environment
will cause re-planning from scratch. Based on A*, various improved algorithms were proposed. One sort of
algorithms is reforming A* into anytime algorithms, which could rapidly give a initial feasible solution and find out
better solutions as the computing time increases. Anytime Repairing A* [35], Restarting Weighted A* [36],
Anytime Window A* [37], Beam-Stack Search [38], ITSA* [39], Anytime Nonparametric A* [40] are of this class.
Another sort is solving the problem of eight-neighborhood limit on grid map. A* with Post-Smoothing algorithm
[41] post-processes the results of A* to get a smoother path. Theta* algorithm [42] searches shortcuts by visibility
check between parent node and successive nodes. The algorithm could find out any-direction path, and has a good
performance on optimality and time efficiency. In addition, there are other improvement ideas. For example, Block
A* algorithm [43] takes the strategy of trading space for time. It operates lots of grids simultaneously by using local
distance database, and gets a faster computation speed than A*.
4.3. Local path planning algorithms
Common local path planning algorithms include Artificial Potential Field (APF) algorithms, Bug-like algorithms,
dynamic heuristic re-planning algorithms and terrain-evaluation-based algorithms.
4.3.1. Artificial Potential Field algorithms
APF [44] virtualizes the rover environment as an artificial potential field, and the repulsions from obstacles and
attraction from target co-affect the rover motion. It is simple, computational efficient and easy for real-time control.
However, APF still has many disadvantages such as local minimum problem, goals non-reachable and oscillations.
Many improved algorithms are proposed to tackle local minimum problem. These works can be classified into four
approaches according to their control strategy [45]. The first kind of approach changes the repulsive potential
functions (including its angle), such as reference [46]. The second kind of approach provides the additional control
force, such as reference [47]. The third kind of approach adds virtual obstacle (or virtual goal) for repelling (or
attracting) the mobile robots from the local minima (or to the target), such as reference [48]. The fourth kind of
approach is to employ intelligent techniques, such as “follow wall” strategy in [49].
4.3.2. Bug-like algorithms
Bug-like algorithms is a kind of simple local path planning algorithms that guarantee of finding local optimal
solutions. In the classical Bug1 and Bug2 algorithms [50], the rover only has two behavior modes: move on a
straight line towards goal and follow a boundary. Because of the simplicity and huge application potential, many
improved algorithms are proposed, including Alg, DistBug, Com, Class1, Rev, TangentBug, Ave, VisBug, OneBug,
LeaveBug, CautiousBug, Bug2+, PointBug, ABug, Ibug, KBug, DHBug, etc [51]. Most of them focus on increasing
efficiency. For example, TangentBug [52] introduces omni-directional range sensing data and searches local
shortcuts to bypass obstacles based on local tangent map. CautiousBug [53] adopts conservative spiral search to
choose the bypassing direction and reduces path length greatly. But these algorithms adopt over-ideal assumptions
such as global positioning and regarding the rover as a point, hence generally they fit for indoor structural
environment but not for planetary surface environment. Some algorithms tried to relax the assumptions. For instance,
IBug [54] regards the target as virtual signal tower and uses signal strength sensor to guide the rover towards
stronger signal with no need of accurate rover position and heading. WedgeBug [55] introduces the virtual rotary
scan action of sensors to reduce the meaningless move of rover, and adopts visual dead-reckoning technology to
enhance autonomous navigation capability. The algorithm is more practical in planetary surface environment and
has been tested on the Rocky 7 prototype.
4.3.3. Dynamic heuristic replanning algorithms
Aiming at the shortcoming of A* that re-plan from the scratch when environment locally changes, a lot of
improved heuristic and incremental algorithms are proposed to get better results in unknown, partially known or
dynamic environments. D* algorithm [56] generalizes A* into dynamic environment. As a re-planning algorithm,
D* only needs revise local changed nodes, hence greatly improves search efficiency. But D* is hard to understand,
analyze, realize and extend. Lifelong Planning A*(LPA*) [57] is the incremental version of A* and is a plan reuse
Wang Qiong/ Procedia Engineering 00 (2014) 000–000
7
algorithm. It improves efficiency and keeps the simplicity and convenience of A* as well. Based on LPA*, D* Lite
[58] was proposed. It is a simplified version of D* and generates the same path as D*, but is simpler and more
extensible. Therefore, subsequent D*-like algorithms are mostly based on D* Lite.
One category of algorithms is solving the problem of eight-neighborhood limit on grid map. Field D* [14] adopts
linear interpolation to eliminate the non-smooth path problem caused by discrete state transfer between adjacent
grids. It was applied on Opportunity, Spirit and Curiosity. SD* Lite [59] checks the visibility between nodes, and
uses Euclid distance to estimate its cost. Incremental Phi* [60] is the incremental version of Basic Theta*, taking
both advantages of any-angle algorithms with optimal path and that of incremental algorithms with high efficiency.
Another category is improving search efficiency. DD* Lite [61] uses dominance relationship between states to
exclude regions that need not be explored, thus improves efficiency in large state space. Improved Dynamic A* [62]
further reduces the re-computation in D* Lite because of terrain changes and remarkably speeds up the computation.
There is also a category of algorithms that synthesizes the advantages of multiple algorithms. Anytime Dynamic A*
[63] integrates the anytime responding feature of ARA* and the incremental re-planning feature of D* Lite.
ROAMAN [3] synthesize the conciseness of environment representation of configuration space method and the high
efficiency of D*. It was tested in FIDO rover prototype.
4.3.4. Terrain-evaluation-based algorithms
The basic idea of terrain-evaluation-based algorithms is carrying out statistic analysis on local terrain around the
rover and then determining the best path. Morphin [64] evaluates the goodness, certainty, traversability and other
measures of each rover-sized terrain patch, generates goodness map and finds the most traversable path among
several possible paths in front of the rover. The algorithm was tested on Ratler and Hyperion prototypes. It could
nicely deal with environment modeling error and positioning error, and works well together with global planning
algorithms such as D*. GESTALT [65] improves Morphin on several areas, including patch shape, computation
method of goodness, merging method of goodness maps, voting standard for forward arcs and so on. The algorithm
is more practical and was used in the local hazard-avoidance module of Opportunity, Spirit and Curiosity.
Probabilistic Terrain Analysis (PTA) [66] uses probabilistic techniques to integrate range measurements over time,
and relies on efficient statistical tests for distinguishing drivable from non-drivable terrain. However it is a binary
detection algorithm providing a simple go/no-go solution. Loh J E et al. [67] proposed a new roughness metric to
create a roughness map and provide planetary rovers with qualitative information about the terrain and difficulty of
traversal. It is computationally efficient, and runs in linear time.
5. Conclusion and outlook
Mission planning technology is the basis and key of safe and efficient roving exploration on planetary surface,
hence much research has been done on it. In order to satisfy the demands of future planetary roving exploration
mission, mission planning technology should be studied further on the following aspects: (1) Mission-level joint
path-behavior planning algorithm under complex constraints. Current mission planning algorithms either focus on
behavior planning too much, or on path planning too much. How to balance the two better and realize optimal
planning of paths and behaviors at the same time, needs further studies. (2) Combination of multi-layer planning
algorithms. The coarse-to-fine, macro-to-micro hierarchical planning structure is a classical structure of robotic
intelligent control. But the next stage of development will be realizing the combination and interaction of multilayer planning algorithms better and improving the effectiveness and optimality of integrated planning results. (3)
High performance local path planning algorithm adaptive for complex terrains. The range of future planetary roving
exploration missions would be much larger and the terrains would be much more complex. Hence, high performance
local path planning algorithms adaptive for complex terrains or tough situations need be further studied.
References
[1] Y. Jia, J.X. Chen, Z.R. Shen, Z. Zhang, Lunar rover, Aeronautical Manufacturing Technology, (2007) 008.(Chinese)
[2] S. Hayati, R. Volpe, P. Backes, J. Balaram, R. Welch, R. Ivlev, G. Tharp, S. Peters, T. Ohm, R. Petras, The Rocky 7 rover: A Mars
sciencecraft prototype, in: Proc. of 1997 Intl. Conf. on Robotics and Automation, IEEE, 1997, pp. 2458-2464.
8
Wang Qiong/ Procedia Engineering00 (2014) 000–000
[3] T. Huntsberger, H. Aghazarian, Y. Cheng, E.T. Baumgartner, E. Tunstel, C. Leger, A. Trebi-Ollennu, P.S. Schenker, Rover autonomy for
long range navigation and science data acquisition on planetary surfaces, in: Proc. of 2002 Intl. Conf. on Robotics and Automation, IEEE,
2002, pp. 3161-3168.
[4] A. Kemurdjian, V. Gromov, V. Mishkinyuk, V. Kucherenko, P. Sologub, Small marsokhod configuration, in: Proc. of 1992 Intl. Conf. on
Robotics and Automation, IEEE, 1992, pp. 165-168.
[5] J. Purvis, P. Klarer, RATLER: Robotic All-Terrain Lunar Exploration Rover, in: Proc. of 6th Annual Workshop on Space Operations
Applications and Research, Vol. 2, 1993, pp. 174-179.
[6] D. Wettergreen, D. Bapna, M. Maimone, G. Thomas, Developing Nomad for robotic exploration of the Atacama Desert, Robotics and
Autonomous Systems, 26 (1999) 127-148.
[7] P. Tompkins, Mission-directed path planning for planetary rover exploration, Carnegie Mellon Univ., Pittsburgh, 2005.
[8] B. Fontaine, D. Termont, L. Steinicke. Autonomous operations of a micro-rover for geo-science on Mars, in: Proc. of 6th ESA workshop on
Advanced Space Technologies for Robotics and Automation, 2000, pp. 128-131.
[9] R. Siegwart, T. Estier, Y. Crausaz. Innovative design for wheeled locomotion in rough terrain, in: Proc. of 6th Intl. Conf. on Intelligent
Autonomous Systems, 2000, pp. 151-162.
[10] SEV concept at: http://www.nasa.gov/exploration/technology/space_exploration_vehicle/index.html
[11] B.H. Wilcox, ATHLETE: A cargo and habitat transporter for the moon, in: Proc. of Aerospace Conf., 2009, pp. 1-7.
[12] B. Imhof, S. Ransom, Pressurized rover for moon and mars surface missions, in: Proc. of 38th Assembly of the Committee on Space
Research, 2010, pp. 437.
[13] Z.W. Wang, G. Guo, Present situation and future development of mobile robot navigation technology, Robot, 25 (2003) 470-474. (Chinese)
[14] M.W. Maimone, P.C. Leger, J.J. Biesiadecki. Overview of the mars exploration rovers’ autonomous mobility and vision capabilities, in:
Proc. of 2007 IEEE Intl. Conf. on Robotics and Automation, 2007.
[15] Y. Xie, Mission planning technology in teleoperation of lunar rovers, Univ. of Science and Technology of China, Hefei, 2011. (Chinese)
[16] B. Cooper. Driving on the surface of Mars using the Rover Control Workstation, in: Proc. of 5th Intl. Conf. on Space Operations, 1998.
[17] M. Ai-Chang, J. Bresina, L. Charest, A. Chase, J.C. Hsu, A. Jonsson, B. Kanefsky, P. Morris, K. Rajan, J. Yglesias, B.G. Chafin, W.C. Dias,
and P.F. Maldague, MAPGEN: Mixed-initiative planning and scheduling for the mars exploration rover mission, IEEE Intelligent Systems,
19 (2004) 8-12.
[18] T. Estlin, F. Fisher, D. Gaines, C. Chouinard, S. Schaffer, I. Nesnas, Continuous planning and execution for an autonomous rover, in: Proc.
of 3rd Intl. NASA Workshop on Planning and Scheduling for Space, 2002.
[19] J.L. Bresina, P.H. Morris. Mission operations planning: Beyond MAPGEN, in: Proc. of 2nd IEEE Intl. Conf. on Space Mission Challenges
for Information Technology, 2006, pp. 151-156.
[20] S. Peng, Y. Jia, Mission-level path planning for lunar rover, Spacecraft Engineering, 19 (2010) 35-42. (Chinese)
[21] P. Song, H.H. Ju, G.C. Dong, H. Li, Mission planning methods of mobile robots based on SD* Lite, in: Proc. of 31st Chinese Control Conf.,
2012. (Chinese)
[22] J.L. Zhou, Y. Xie, Q. Zhang, K. Wu, Y. Xia, Z. Zhang, Research on mission planning in teleoperation of lunar rovers, Science China
Information Sciences, 44 (2014) 693-693. (Chinese)
[23] P. Song, Research on mission planning method of lunar rover based on predicate logic, Beijing Univ. of Technology, Beijing, 2012. (Chinese)
[24] N. Nilsson, A mobile automaton: An application of artificial intelligence techniques, in: Proc. of 1st Intl. Joint Conf. on Artificial
Intelligence, 1969, pp. 509-520.
[25] Y.H. Liu, S. Arimoto, Proposal of tangent graph and extended tangent graph for path planning of mobile robots, in: Proc. of 1991 IEEE Intl.
Conf. on Robotics and Automation, 1991, pp. 312-317.
[26] M. Sharir, C.K. Yap, Retraction: A new approach to motion-planning, in: Proc. of 15th Annual ACM Symp. on Theory of computing, 1983,
pp. 207-220.
[27] M.K. Habib, H. Asama, Efficient method to generate collision free path for autonomous mobile robot based on new free space structuring
approach, in: Proc. of 4th IEEE/RSJ Intl. Conf. on Intelligent Robots and Systems, 1991, pp. 563-567.
[28] D.Q. Zhu, M.Z. Yan, Survey on technology of mobile robot path planning, Control and Decision, 25 (2010) 961-967. (Chinese)
[29] S. Kambhampati, L. Davis, Multiresolution path planning for mobile robots, IEEE J. of Robotics and Automation, 2 (1986) 135-145.
[30] R.J. Szczerba, D.Z. Chen, Using framed-subspaces to solve the 2-D and 3-D weighted region problem, Notre Dame Univ., Notre Dame, 1996.
[31] S. Behnke, Local multiresolution path planning, in Proc. of 7th RoboCup Intl. Symp., Springer, 2004, pp. 332-343.
[32] J.Y. Hwang, J.S. Kim, S.S. Lim, K.H. Park, A fast path planning by path graph optimization, IEEE Trans. on Systems, Man and Cybernetics,
Part A: Systems and Humans, 33 (2003) 121-129.
[33] E.W. Dijkstra, A note on two problems in connection with graphs, Numerische mathematik, 1 (1959) 269-271.
[34] N.J. Nilsson, Principles of artificial intelligence, Morgan Kaufmann, Burlington, 1980.
[35] M. Likhachev, G. Gordon, S. Thrun, ARA*: Anytime A* with provable bounds on sub-optimality, in: Proc. of 17th Annual Conf. on Neural
Information Processing Systems, 2003.
[36] S. Richter, J.T. Thayer, W. Ruml, The joy of forgetting: faster anytime search via restarting, in: Proc. of 20th Intl. Conf. on Automated
Planning and Scheduling, 2010, pp. 137-144.
[37] S. Aine, P.P. Chakrabarti, R. Kumar, AWA*-a window constrained anytime heuristic search algorithm, in: Proc. of 20th Intl. Joint Conf. on
Artificial Intelligence, 2007, pp. 2250-2255.
[38] R. Zhou, E.A. Hansen, Beam-stack search: Integrating backtracking with beam search, in: Proc. of 15th Intl. Conf. on Automated Planning
and Scheduling, 2005, pp. 90-98.
Wang Qiong/ Procedia Engineering 00 (2014) 000–000
9
[39] D.A. Furcy, ITSA*: Iterative tunneling search with A*, in: Proc. of AAAI Workshop on Heuristic Search, Memory-Based Heuristics and
their Applications, 2006, pp. 21-26.
[40] J. van den Berg, R. Shah, A. Huang, K. Goldberg, ANA*: Anytime nonparametric A*, in Proc. of 25th AAAI Conf. on Artificial
Intelligence, 2011, pp. 105-111.
[41] J.P. Gonzalez, A. Stentz, Planning with uncertainty in position: An optimal planner, Carnegie Mellon Univ., Pittsburgh, 2004.
[42] K. Daniel, A. Nash, S. Koenig, A. Felner, Theta*: Any-angle path planning on grids, J. of Artificial Intelligence Research, 39 (2010) 533-579.
[43] P. Yap, N. Burch, R. Holte, J. Schaeffer, Block A*: Database-driven search with applications in any-angle path-planning, in: Proc. of 25th
AAAI Conf. on Artificial Intelligence, 2011, pp. 120-125.
[44] O. Khatib. Real-time obstacle avoidance for maniplators and mobile robots, Intl. J. of Robotics Research, 5 (1986) 90–98.
[45] Q. Li, L. Wang, B. Chen, Z. Zhou, An improved artificial potential field method for solving local minimum problem, in: Proc. of 2nd Intl.
Conf. on Intelligent Control and Information Processing, 2011, pp. 420-424.
[46] S.S. Ge, Y.J. Cui, New potential functions for mobile robot path planning, IEEE Trans. on Robotics and Automation, 16 (2000) 615-620.
[47] L. Tang, S. Dian, G. Gu, K. Zhou, S. Wang, X. Feng, A novel potential field method for obstacle avoidance and path planning of mobile
robot, in: Proc. of 3rd Intl. Conf. on Computer Science and Information Technology, IEEE, 2010, pp. 633-637.
[48] M.C. Lee, M.G. Park, Artificial potential field based path planning for mobile robots using a virtual obstacle concept, in: Proc. of 2003
IEEE/ASME Intl. Conf. on Advanced Intelligent Mechatronics, 2003, pp. 735-740.
[49] J. Antich, A. Ortiz, Bug-based T2: A new globally convergent potential field approach to obstacle avoidance, in: Proc. of 2006 IEEE/RSJ Intl.
Conf. on Intelligent Robots and Systems, 2006, pp. 430-435.
[50] V.J. Lumelsky, A.A. Stepanov, Path-planning strategies for a point mobile automaton moving amidst unknown obstacles of arbitrary shape,
Algorithmica, 2 (1987) 403-430.
[51] J. Ng, An analysis of mobile robot navigation algorithms in unknown environments, Univ. of Western Australia, Perth, 2010.
[52] I. Kamon, E. Rivlin, E. Rimon, A new range-sensor based globally convergent navigation algorithm for mobile robots, in: Proc. of 1996
IEEE Intl. Conf. on Robotics and Automation, 1996, pp. 429-435.
[53] E. Magid, E. Rivlin. CautiousBug: A competitive algorithm for sensory-based robot navigation, in: Proc. of 2004 Intl. Conf. on Intelligent
Robots and Systems, IEEE/RSJ, 2004, pp. 2757-2762.
[54] K. Taylor, S.M. LaValle, I-Bug: An intensity-based bug algorithm, in: Proc. of 2009 IEEE Intl. Conf. on Robotics and Automation, 2009, pp.
3981-3986.
[55] S.L. Laubach, J.W. Burdick, An autonomous sensor-based path-planner for planetary microrovers, in: Proc. of 1999 IEEE Intl. Conf. on
Robotics and Automation, 1999, pp. 347-354.
[56] A. Stentz, Optimal and efficient path planning for partially-known environments, in: Proc. of Intl. J. of Robotics and Automation, 1993, pp.
3310-3317.
[57] S. Koenig, M. Likhachev, D. Furcy, Lifelong Planning A*, Artificial Intelligence, 155 (2004) 93-146.
[58] S. Koenig, M. Likhachev, Improved fast replanning for robot navigation in unknown terrain, in: Proc. of 2002 IEEE Intl. Conf. on Robotics
and Automation, 2002, pp. 968-975.
[59] S. Dong, H. Ju, H. Xu, An improvement of D* lite algorithm for planetary rover mission planning, in: Proc. of 2011 Intl. Conf. on
Mechatronics and Automation, 2011, pp. 1810-1815.
[60] A. Nash, S. Koenig, M. Likhachev, Incremental Phi*: Incremental any-angle path planning on grids, in: Proc. of 21st Intl. Joint Conf. on
Artificial Intelligence, 2009, pp. 1824-1830.
[61] Mills-Tettey G A, A. Stentz, M.B. Dias, DD* Lite: Efficient incremental search with state dominance, in: Proc. of 21st National Conf. on
Artificial Intelligence, 2006, pp. 1032-1038.
[62] W. Yue, J. Franco, W. Cao, H. Yue, ID* Lite: Improved D* Lite algorithm, in: Proc. of 26th Symp. on Applied Computing, 2011, pp. 1364-1369.
[63] M. Likhachev, D.I. Ferguson, G.J. Gordon, A. Stenz, S. Thrun, Anytime dynamic A*: An anytime, replanning algorithm, in: Proc. of 2005
Intl. Conf. on Automated Planning and Scheduling, 2005, pp. 262-271.
[64] S. Singh, R. Simmons, T. Smith, A. Stentz, V. Verma, A. Yahja, K. Schwehr, Recent progress in local and global traversability for planetary
rovers, in: Proc. of 2000 IEEE Intl. Conf. on Robotics and Automation, 2000, pp. 1194-1200.
[65] S.B. Goldberg, M.W. Maimone, L. Matthies, Stereo vision and rover navigation software for planetary exploration, in: Proc. of 2002 IEEE
Aerospace Conf., 2002, pp. 2025-2036.
[66] S. Thrun, M. Montemerlo, A. Aron, Probabilistic terrain analysis for high-speed desert driving, in: Proc. of 2006 Robotics: Science and
Systems Conf., 2006, pp. 16-19.
[67] J.E. Loh, G.H. Elkaim, R.E. Curry, Roughness map for autonomous rovers, in: Proc. of 2013 American Control Conf., 2013.
Download