Asymptotically Optimal Fast Marching Tree algorithm developed by L. Janson and M. Pavone. More...
#include <ompl/geometric/planners/fmt/FMT.h>
Classes  
struct  CostIndexCompare 
class  Motion 
Representation of a motion. More...  
struct  MotionCompare 
Comparator used to order motions in a binary heap. More...  
Public Member Functions  
FMT (const base::SpaceInformationPtr &si)  
virtual void  setup () 
Perform extra configuration steps, if needed. This call will also issue a call to ompl::base::SpaceInformation::setup() if needed. This must be called before solving.  
virtual base::PlannerStatus  solve (const base::PlannerTerminationCondition &ptc) 
Function that can solve the motion planning problem. This function can be called multiple times on the same problem, without calling clear() in between. This allows the planner to continue work for more time on an unsolved problem, for example. If this option is used, it is assumed the problem definition is not changed (unpredictable results otherwise). The only change in the problem definition that is accounted for is the addition of starting or goal states (but not changing previously added start/goal states). The function terminates if the call to ptc returns true. More...  
virtual void  clear () 
Clear all internal datastructures. Planner settings are not affected. Subsequent calls to solve() will ignore all previous work.  
virtual void  getPlannerData (base::PlannerData &data) const 
Get information about the current run of the motion planner. Repeated calls to this function will update data (only additions are made). This is useful to see what changed in the exploration datastructure, between calls to solve(), for example (without calling clear() in between).  
void  setNumSamples (const unsigned int numSamples) 
Set the number of states that the planner should sample. The planner will sample this number of states in addition to the initial states. If any of the goal states are not reachable from the randomly sampled states, those goal states will also be added. The default value is 1000.  
unsigned int  getNumSamples () const 
Get the number of states that the planner will sample.  
void  setNearestK (bool nearestK) 
If nearestK is true, FMT will be run using the Knearest strategy.  
bool  getNearestK () const 
Get the state of the nearestK strategy.  
void  setRadiusMultiplier (const double radiusMultiplier) 
The planner searches for neighbors of a node within a cost r, where r is the value described for FMT* in Section 4 of L. Janson, E. Schmerling, A. Clark, M. Pavone. Fast marching tree: a fast marching samplingbased method for optimal motion planning in many dimensions. The International Journal of Robotics Research, 34(7):883921, 2015. For guaranteed asymptotic convergence, the user should choose a constant multiplier for the search radius that is greater than one. The default value is 1.1. In general, a radius multiplier between 0.9 and 5 appears to perform the best.  
double  getRadiusMultiplier () const 
Get the multiplier used for the nearest neighbors search radius.  
void  setFreeSpaceVolume (const double freeSpaceVolume) 
Store the volume of the obstaclefree configuration space. If no value is specified, the default assumes an obstaclefree unit hypercube, freeSpaceVolume = (maximumExtent/sqrt(dimension))^(dimension)  
double  getFreeSpaceVolume () const 
Get the volume of the free configuration space that is being used by the planner.  
void  setCacheCC (bool ccc) 
Sets the collision check caching to save calls to the collision checker with slightly memory usage as a counterpart.  
bool  getCacheCC () const 
Get the state of the collision check caching.  
void  setHeuristics (bool h) 
Activates the cost to go heuristics when ordering the heap.  
bool  getHeuristics () const 
Returns true if the heap is ordered taking into account cost to go heuristics.  
void  setExtendedFMT (bool e) 
Activates the extended FMT*: adding new samples if planner does not finish successfully.  
bool  getExtendedFMT () const 
Returns true if the extended FMT* is activated.  
Public Member Functions inherited from ompl::base::Planner  
Planner (const Planner &)=delete  
Planner &  operator= (const Planner &)=delete 
Planner (const SpaceInformationPtr &si, const std::string &name)  
Constructor.  
virtual  ~Planner () 
Destructor.  
template<class T >  
T *  as () 
Cast this instance to a desired type. More...  
template<class T >  
const T *  as () const 
Cast this instance to a desired type. More...  
const SpaceInformationPtr &  getSpaceInformation () const 
Get the space information this planner is using.  
const ProblemDefinitionPtr &  getProblemDefinition () const 
Get the problem definition the planner is trying to solve.  
const PlannerInputStates &  getPlannerInputStates () const 
Get the planner input states.  
virtual void  setProblemDefinition (const ProblemDefinitionPtr &pdef) 
Set the problem definition for the planner. The problem needs to be set before calling solve(). Note: If this problem definition replaces a previous one, it may also be necessary to call clear().  
PlannerStatus  solve (const PlannerTerminationConditionFn &ptc, double checkInterval) 
Same as above except the termination condition is only evaluated at a specified interval.  
PlannerStatus  solve (double solveTime) 
Same as above except the termination condition is solely a time limit: the number of seconds the algorithm is allowed to spend planning.  
const std::string &  getName () const 
Get the name of the planner.  
void  setName (const std::string &name) 
Set the name of the planner.  
const PlannerSpecs &  getSpecs () const 
Return the specifications (capabilities of this planner)  
virtual void  checkValidity () 
Check to see if the planner is in a working state (setup has been called, a goal was set, the input states seem to be in order). In case of error, this function throws an exception.  
bool  isSetup () const 
Check if setup() was called for this planner.  
ParamSet &  params () 
Get the parameters for this planner.  
const ParamSet &  params () const 
Get the parameters for this planner.  
const PlannerProgressProperties &  getPlannerProgressProperties () const 
Retrieve a planner's planner progress property map.  
virtual void  printProperties (std::ostream &out) const 
Print properties of the motion planner.  
virtual void  printSettings (std::ostream &out) const 
Print information about the motion planner's settings.  
Protected Types  
typedef ompl::BinaryHeap< Motion *, MotionCompare >  MotionBinHeap 
A binary heap for storing explored motions in costtocome sorted order.  
Protected Member Functions  
double  distanceFunction (const Motion *a, const Motion *b) const 
Compute the distance between two motions as the cost between their contained states. Note that for computationally intensive cost functions, the cost between motions should be stored to avoid duplicate calculations.  
void  freeMemory () 
Free the memory allocated by this planner.  
void  sampleFree (const ompl::base::PlannerTerminationCondition &ptc) 
Sample a state from the free configuration space and save it into the nearest neighbors data structure.  
void  assureGoalIsSampled (const ompl::base::GoalSampleableRegion *goal) 
For each goal region, check to see if any of the sampled states fall within that region. If not, add a goal state from that region directly into the set of vertices. In this way, FMT is able to find a solution, if one exists. If no sampled nodes are within a goal region, there would be no way for the algorithm to successfully find a path to that region.  
double  calculateUnitBallVolume (const unsigned int dimension) const 
Compute the volume of the unit ball in a given dimension.  
double  calculateRadius (unsigned int dimension, unsigned int n) const 
Calculate the radius to use for nearest neighbor searches, using the bound given in L. Janson, E. Schmerling, A. Clark, M. Pavone. Fast marching tree: a fast marching samplingbased method for optimal motion planning in many dimensions. The International Journal of Robotics Research, 34(7):883921, 2015. The radius depends on the radiusMultiplier parameter, the volume of the free configuration space, the volume of the unit ball in the current dimension, and the number of nodes in the graph.  
void  saveNeighborhood (Motion *m) 
Save the neighbors within a neighborhood of a given state. The strategy used (nearestK or nearestR depends on the planner configuration.  
void  traceSolutionPathThroughTree (Motion *goalMotion) 
Trace the path from a goal state back to the start state and save the result as a solution in the Problem Definiton.  
bool  expandTreeFromNode (Motion **z) 
Complete one iteration of the main loop of the FMT* algorithm: Find K nearest nodes in set Unvisited (or within a radius r) of the node z. Attempt to connect them to their optimal costtocome parent in set Open. Remove all newly connected nodes fromUnvisited and insert them into Open. Remove motion z from Open, and update z to be the current lowest costtocome node in Open.  
void  updateNeighborhood (Motion *m, const std::vector< Motion * > nbh) 
For a motion m, updates the stored neighborhoods of all its neighbors by by inserting m (maintaining the costbased sorting). Computes the nearest neighbors if there is no stored neighborhood.  
Motion *  getBestParent (Motion *m, std::vector< Motion * > &neighbors, base::Cost &cMin) 
Returns the best parent and the connection cost in the neighborhood of a motion m.  
Protected Member Functions inherited from ompl::base::Planner  
template<typename T , typename PlannerType , typename SetterType , typename GetterType >  
void  declareParam (const std::string &name, const PlannerType &planner, const SetterType &setter, const GetterType &getter, const std::string &rangeSuggestion="") 
This function declares a parameter for this planner instance, and specifies the setter and getter functions.  
template<typename T , typename PlannerType , typename SetterType >  
void  declareParam (const std::string &name, const PlannerType &planner, const SetterType &setter, const std::string &rangeSuggestion="") 
This function declares a parameter for this planner instance, and specifies the setter function.  
void  addPlannerProgressProperty (const std::string &progressPropertyName, const PlannerProgressProperty &prop) 
Add a planner progress property called progressPropertyName with a property querying function prop to this planner's progress property map.  
Protected Attributes  
MotionBinHeap  Open_ 
A binary heap for storing explored motions in costtocome sorted order. The motions in Open have been explored, yet are still close enough to the frontier of the explored set Open to be connected to nodes in the unexplored set Unvisited.  
std::map< Motion *, std::vector< Motion * > >  neighborhoods_ 
A map linking a motion to all of the motions within a distance r of that motion.  
unsigned int  numSamples_ 
The number of samples to use when planning.  
unsigned int  collisionChecks_ 
Number of collision checks performed by the algorithm.  
bool  nearestK_ 
Flag to activate the K nearest neighbors strategy.  
bool  cacheCC_ 
Flag to activate the collision check caching.  
bool  heuristics_ 
Flag to activate the cost to go heuristics.  
double  NNr_ 
Radius employed in the nearestR strategy.  
unsigned int  NNk_ 
K used in the nearestK strategy.  
double  freeSpaceVolume_ 
The volume of the free configuration space, computed as an upper bound with 95% confidence.  
double  radiusMultiplier_ 
This planner uses a nearest neighbor search radius proportional to the lower bound for optimality derived for FMT* in Section 4 of L. Janson, E. Schmerling, A. Clark, M. Pavone. Fast marching tree: a fast marching samplingbased method for optimal motion planning in many dimensions. The International Journal of Robotics Research, 34(7):883921, 2015. The radius multiplier is the multiplier for the lower bound. For guaranteed asymptotic convergence, the user should choose a multiplier for the search radius that is greater than one. The default value is 1.1. In general, a radius between 0.9 and 5 appears to perform the best.  
std::shared_ptr< NearestNeighbors< Motion * > >  nn_ 
A nearestneighbor datastructure containing the set of all motions.  
base::StateSamplerPtr  sampler_ 
State sampler.  
base::OptimizationObjectivePtr  opt_ 
The cost objective function.  
Motion *  lastGoalMotion_ 
The most recent goal motion. Used for PlannerData computation.  
base::State *  goalState_ 
Goal state caching to accelerate cost to go heuristic computation.  
bool  extendedFMT_ 
Add new samples if the tree was not able to find a solution.  
Protected Attributes inherited from ompl::base::Planner  
SpaceInformationPtr  si_ 
The space information for which planning is done.  
ProblemDefinitionPtr  pdef_ 
The user set problem definition.  
PlannerInputStates  pis_ 
Utility class to extract valid input states.  
std::string  name_ 
The name of this planner.  
PlannerSpecs  specs_ 
The specifications of the planner (its capabilities)  
ParamSet  params_ 
A map from parameter names to parameter instances for this planner. This field is populated by the declareParam() function.  
PlannerProgressProperties  plannerProgressProperties_ 
A mapping between this planner's progress property names and the functions used for querying those progress properties.  
bool  setup_ 
Flag indicating whether setup() has been called.  
Additional Inherited Members  
Public Types inherited from ompl::base::Planner  
typedef std::function< std::string()>  PlannerProgressProperty 
Definition of a function which returns a property about the planner's progress that can be queried by a benchmarking routine.  
typedef std::map< std::string, PlannerProgressProperty >  PlannerProgressProperties 
A dictionary which maps the name of a progress property to the function to be used for querying that property.  
Detailed Description
Asymptotically Optimal Fast Marching Tree algorithm developed by L. Janson and M. Pavone.
 Short description
 FMT* is an asymptoticallyoptimal samplingbased motion planning algorithm, which is guaranteed to converge to a shortest path solution. The algorithm is specifically aimed at solving complex motion planning problems in highdimensional configuration spaces. The FMT* algorithm essentially performs a lazy dynamic programming recursion on a set of probabilisticallydrawn samples to grow a tree of paths, which moves steadily outward in costtocome space.
 Deviation from the paper
 The implementation includes a cache in the collision checking since the original algorithm could check the same collision more than once. It increases the memory requirements to O(n logn), but as samples tend to infinity this bound tend to O(n).
It also implements the resampling strategy (extended FMT) included in the BiDirectional FMT* paper.
 External documentation
 L. Janson, E. Schmerling, A. Clark, M. Pavone. Fast marching tree: a fast marching samplingbased method for optimal motion planning in many dimensions. The International Journal of Robotics Research, 34(7):883921, 2015. DOI: 10.1177/0278364915577958
[PDF]
J. A. Starek, J. V. Gomez, E. Schmerling, L. Janson, L. Moreno, and M. Pavone, An AsymptoticallyOptimal SamplingBased Algorithm for Bidirectional Motion Planning, in IEEE/RSJ International Conference on Intelligent Robots Systems, 2015. [PDF]
Member Function Documentation

virtual 
Function that can solve the motion planning problem. This function can be called multiple times on the same problem, without calling clear() in between. This allows the planner to continue work for more time on an unsolved problem, for example. If this option is used, it is assumed the problem definition is not changed (unpredictable results otherwise). The only change in the problem definition that is accounted for is the addition of starting or goal states (but not changing previously added start/goal states). The function terminates if the call to ptc returns true.
 Todo:
 Create a PRMlike connection strategy
Implements ompl::base::Planner.
The documentation for this class was generated from the following files: