pmap.h File Reference

The pmap library performs simple particle-filter-based map building (SLAM) for a single robot. Generally speaking, the output maps are topologically correct, but a little rough around the edges (some minor scan mis-alignments). For very precise results, the output maps should be refined using the rmap library. More...

#include "gsl/gsl_rng.h"
#include "slap.h"

Go to the source code of this file.

Classes
struct	pmap_nbor_t
	Structure for neighborhood lookup table. More...
struct	pmap_scan_t
	Structure storing a single range scan. More...
struct	pmap_sample_t
	Structure describing a single sample. More...
struct	pmap_t
	Structure describing the map. More...
Defines
#define	PMAP_MAX_RANGES   401
	Limits.
#define	PMAP_GET_SAMPLE(self, i)   (self->samples + self->sample_set * self->samples_len + (i))
	Sample access macros.
#define	PMAP_GRIDX(self, x)   ((int) floor((x) / self->grid_res) + self->grid_sx / 2)
	Grid access macros.
Functions
pmap_t *	pmap_alloc (int num_ranges, double range_max, double range_start, double range_step, int samples_len, double grid_width, double grid_height, double grid_scale)
	Allocate object.
void	pmap_free (pmap_t *self)
	Free object.
void	pmap_init_nbors (pmap_t *self)
	Create neighborhood LUT.
void	pmap_set_pose (pmap_t *self, pose2_t pose)
	Set the initial robot pose. This must be called before the first call to pmap_update().
void	pmap_update (pmap_t *self, double dtime, pose2_t odom_pose, int num_ranges, double *ranges)
	Update map.
void	pmap_apply_action (pmap_t *self, pose2_t delta)
	Apply the action model to the current sample set.
void	pmap_add_scan (pmap_t *self, double dtime, double *ranges)
	Add a scan to list.
void	pmap_add_scan_sample (pmap_t *self, int sample_index, double *ranges)
void	pmap_apply_sensor (pmap_t *self, double *ranges)
	Apply sensor model to the current sample set.
void	pmap_apply_sensor_sample (pmap_t *self, int sample_index, double *ranges)
	Apply sensor model to a particular sample.
double	pmap_entropy (pmap_t *self, int scan_count)
	Compute entropy.
void	pmap_resample (pmap_t *self, int scan_count)
	Resample.
void	pmap_draw_scan (pmap_t *self, double *ranges)
	Draw the current range scan.
void	pmap_draw_samples (pmap_t *self)
	Draw all samples.
void	pmap_draw_sample (pmap_t *self, int sample_index)
	Draw a particular sample.
void	pmap_draw_map (pmap_t *self, double scale)
	Draw the current best map.
void	pmap_draw_sample_map (pmap_t *self, double scale, int sample_index)
	Draw a particular sample map.

---

Detailed Description

The pmap library performs simple particle-filter-based map building (SLAM) for a single robot. Generally speaking, the output maps are topologically correct, but a little rough around the edges (some minor scan mis-alignments). For very precise results, the output maps should be refined using the rmap library.

How it works

The library maintains a PF over possible maps (i.e., each particle, or sample, represents a complete map). Laser and odometry data is used to incrementally update the filter; resampling is used to concentrate particles around likely maps.

Caveats

• The space of possible maps is vast, and the particle set necessarily represents a very sparse sampling of this space. In practice, it therefore necessary to pre-process raw odometry data (using the lodo library, for example) to minimize the odometric drift rate.

• Maintaining a PF over maps is very memory intensive (the pmap library stores a complete map for each particle). For example, a 2500 sq. m map with 10cm resolution requires 0.25 Mb of storage for each map, or 250 Mb of storage for 1000 particles. When using this library, take care not to exceed the physical memory of the machine.

• The algorithm has constant update time for each new sensor reading, but this value scales linearly with the number of particles in the filter. With more than 1000 particles, the algorithm can be very slow.

---

Function Documentation

pmap_t* pmap_alloc (  int  num_ranges, double  range_max, double  range_start, double  range_step, int  samples_len, double  grid_width, double  grid_height, double  grid_scale )

Allocate object. Parameters: num_ranges  Number of range readings in each scan (should be 181). range_max  Maximum useable range value (e.g., 8.00 or 16.00). range_start  Starting angle for range readings (should be -M_PI / 2). range_step  Angular step size for each successive range reading. samples_len  Number of samples in filter. grid_width  Width of occupancy grid (cells). grid_height  Heigt of occupancy grid (cells). grid_scale  Size of each grid cell (m). (should be M_PI / 180).

void pmap_init_nbors (  pmap_t *  self  )

Create neighborhood LUT. For internal use only.

void pmap_apply_action (  pmap_t *  self, pose2_t  delta )

Apply the action model to the current sample set. For internal use only.

void pmap_add_scan (  pmap_t *  self, double  dtime, double *  ranges )

Add a scan to list. For internal use only.

void pmap_add_scan_sample (  pmap_t *  self, int  sample_index, double *  ranges )

Add a scan to a particular sample For internal use only.

void pmap_apply_sensor (  pmap_t *  self, double *  ranges )

Apply sensor model to the current sample set. For internal use only.

void pmap_apply_sensor_sample (  pmap_t *  self, int  sample_index, double *  ranges )

Apply sensor model to a particular sample. For internal use only.

double pmap_entropy (  pmap_t *  self, int  scan_count )

Compute entropy. For internal use only.

void pmap_resample (  pmap_t *  self, int  scan_count )

Resample. For internal use only.

void pmap_draw_map (  pmap_t *  self, double  scale )

Draw the current best map. Parameters: scale  Pixel scale (m/pixel).

void pmap_draw_sample_map (  pmap_t *  self, double  scale, int  sample_index )

Draw a particular sample map. Parameters: scale  Pixel scale (m/pixel).

---