🎾D.7 Smac Planner
Smac Planner plugin မှာ three A* base planning algorithms များဖြစ်တဲ့ 2D A*, Hybrid-A* နဲ့ State Lattice path planners တို့ကို ပါဝင် ပါတယ်။ အတိအကျပြောရမယ်ဆိုရင်တော့ အောက်ကကောင်တွေပါ။
Smac 2D Planner
Smac Hybrid-A* Planner
Smac State Lattice Planner
Nav2_smac_planner package မှာ robot platforms အမျိုးမျိုးအတွက် အဆင့်မြင့်ထားတဲ့ A*-based search algorithm ပါဝင်ပါတယ်။ template node types တွေကိုအသုံးပြုထားပြီး နောက်ထပ် planners အသစ်တွေ develop လုပ်လို့လည်းရပါတယ်။
circular differential-drive robots နဲ့ omni-directional drive robots တွေအတွက်ဆိုရင် SmacPlanner2D planner အသုံးပြုရမှာဖြစ်ပါတယ်။
ကားပုံစံ ackermann နဲ့ legged vehicles robot တွေအတွက်ဆိုရင်တော့ SmacPlannerHybrid plugin ကို support လုပ်ပေးပါတယ်။
Non-circular တွေ၊ arbitrary shaped robot တွေနဲ့ စိတ်ကြိုက် model vehicles များကိိုအသုံးပြုမယ်ဆိုရင်တော့ SmacPlannerLattice plugin ကိုသုံးနိုင်ပါတယ်။ ဒီကောင်သည် omni, diff, ackermann, legged, custom အမျိုးအစားတွေအားလုံးအတွက် support ပေးတယ်လို့ဆိုပါတယ်။
အပေါ်မှာ ဖေါ်ပြခဲ့တဲ့ plugin ၃ထဲက နောက်ဆုံး plugin နှစ်ခုကတော့ kinematically feasible ဖြစ်ပါတယ်။
အောက်မှာ planner သုံးခု နဲ့ ၇၅ မီတာအကို အကြမ်းအားဖြင့် plan လုပ်ရာမှာ ကြာချိန်အသီးသီးကို ဖေါ်ပြထားပါတယ်။
Hybrid-A* computed the path in 144 ms
State Lattice computed the path in 113 ms
2D A* computed the path in 243 ms
For reference: NavFn compute the path in 146 ms, including some nasty path discontinuity artifacts
တချို့ environment များအတွက် 100ms အောက်ကို ပုံမှန် planning လုပ်ပြီး၊ တခါတရံ 200 ms နဲ့ ထိကြာပါတယ်။
Smac 2D Planner
<name>.tolerance
Tolerance in meters between requested goal pose and end of path.
double
0.125
<name>.downsample_costmap
Whether to downsample costmap to another resolution for search.
bool
FALSE
<name>.downsampling_factor
Multiplier factor to downsample costmap by (e.g. if 5cm costmap at 2 downsample_factor, 10cm output).
int
1
<name> .allow_unknown
Whether to allow traversing/search in unknown space.
bool
TRUE
<name>.max_iterations
Maximum number of search iterations before failing to limit compute time, disabled by -1.
int
1000000
<name>.max_on_approach_iterations
Maximum number of iterations after the search is within tolerance before returning approximate path with best heuristic if exact path is not found.
int
1000
<name> .max_planning_time
Maximum planning time in seconds.
double
2
နမူနာ
Smac Hybrid-A* Planner
<name>.downsample_costmap
Whether to downsample costmap to another resolution for search.
bool
FALSE
<name>.downsampling_factor
Multiplier factor to downsample costmap by (e.g. if 5cm costmap at 2 downsample_factor, 10cm output).
int
1
<name>.allow_unknown
Whether to allow traversing/search in unknown space.
bool
TRUE
<name> .tolerance
If an exact path cannot be found, the tolerance (as measured by the heuristic cost-to-goal) that would be acceptable to diverge from the requested pose.
double
0.25
<name> .max_iterations
Maximum number of search iterations before failing to limit compute time, disabled by -1.
int
1000000
<name> .max_on_approach_iterations
Maximum number of iterations once a visited node is within the goal tolerances to continue to try to find an exact match before returning the best path solution within tolerances. Negative values convert to infinite.
int
1000
<name> .max_planning_time
Maximum planning time in seconds.
double
5.0
<name> .analytic_expansion_ratio
Planner will attempt to complete an analytic expansions in a frequency proportional to this value and the minimum heuristic.
double
3.5
<name> .analytic_expansion_max_length
If the length is too far, reject this expansion. This prevents shortcutting of search with its penalty functions far out from the goal itself (e.g. so we don’t reverse half-way across open maps or cut through high cost zones). This should never be smaller than 4-5x the minimum turning radius being used, or planning times will begin to spike.
double
3.0
<name> .motion_model_for_search
Motion model enum string to search with. For Hybrid-A* node, default is “DUBIN”. Options for SE2 are DUBIN or REEDS_SHEPP.
string
“DUBIN”
<name> .angle_quantization_bins
Number of angular bins to use for SE2 search. This can be any even number, but a good baseline is 64 or 72 (for 5 degree increments).
int
72
<name> .minimum_turning_radius
Minimum turning radius in meters of vehicle. Also used in the smoother to compute maximum curvature.
double
0.4
<name> .reverse_penalty
Heuristic penalty to apply to SE2 node if searching in reverse direction. Only used in REEDS_SHEPP motion model.
double
2.0
<name> .change_penalty
Heuristic penalty to apply to SE2 node if changing direction (e.g. left to right) in search. Disabled by default after change to guarantee admissibility of the Hybrid-A* planner.
double
0
<name> .non_straight_penalty
Heuristic penalty to apply to SE2 node if searching in non-straight direction.
double
1.2
<name> .cost_penalty
Heuristic penalty to apply to SE2 node for cost at pose. Allows Hybrid-A* to be cost aware.
double
2
<name> .retrospective_penalty
Heuristic penalty to apply to SE2 node penalty. Causes Hybrid-A* to prefer later maneuvers before earlier ones along the path. Saves search time since earlier (shorter) branches are not expanded until it is necessary. Must be >= 0.0 and <= 1.0. Must be 0.0 to be fully admissible.
double
0.015
<name> .lookup_table_size
Size of the dubin/reeds-sheep distance window to cache, in meters.
double
20.0
<name> .viz_expansions
Whether to publish expansions on the /expansions topic as an array of poses (the orientation has no meaning). WARNING: heavy to compute and to display, for debug only as it degrades the performance.
bool
FALSE
<name> .cache_obstacle_heuristic
Cache the obstacle map dynamic programming distance expansion heuristic between subsiquent replannings of the same goal location. Dramatically speeds up replanning performance (40x) if costmap is largely static.
bool
FALSE
<name> .smooth_path
If true, does simple and fast smoothing post-processing to the path from search
bool
TRUE
<name> .smoother.max_iterations
The maximum number of iterations the smoother has to smooth the path, to bound potential computation.
int
1000
<name> .smoother.w_smooth
Weight for smoother to apply to smooth out the data points
double
0.3
<name> .smoother.w_data
Weight for smoother to apply to retain original data information
double
0.2
<name> .smoother.tolerance
Parameter tolerance change amount to terminate smoothing session
double
1E-10
<name> .smoother.do_refinement
Performs extra refinement smoothing runs. Essentially, this recursively calls the smoother using the output from the last smoothing cycle to further smooth the path for macro-trends. This typically improves quality especially in the Hybrid-A* planner due to the extra “wobbling” it can have due to the very small primitive lengths but may cause the path to get slightly closer to some obstacles.
bool
TRUE
<name> .smoother.refinement_num
Number of times to recursively attempt to smooth, must be >= 1.
int
2
နမူနာ
Smac State Lattice Planner
Note: State Lattice does not have the costmap downsampler due to the minimum control sets being tied with map resolutions on generation. The minimum turning radius is also not a parameter in State Lattice since this was specified at the minimum control set pre-computation phase. See the Smac Planner package to generate custom control sets for your vehicle or use one of our pre-generated examples.
The image above you can see the reverse expansion enabled, such that the robot can back into a tight requested spot close to an obstacle.
<name> .allow_unknown
Whether to allow traversing/search in unknown space.
bool
TRUE
<name> .tolerance
If an exact path cannot be found, the tolerance (as measured by the heuristic cost-to-goal) that would be acceptable to diverge from the requested pose in distance-to-goal.
double
0.25
<name> .max_iterations
Maximum number of search iterations before failing to limit compute time, disabled by -1.
int
1000000
<name> .max_on_approach_iterations
Maximum number of iterations once a visited node is within the goal tolerances to continue to try to find an exact match before returning the best path solution within tolerances.
int
1000
<name> .max_planning_time
Maximum planning time in seconds.
double
5.0
<name> .analytic_expansion_ratio
SE2 node will attempt to complete an analytic expansion with frequency proportional to this value and the minimum heuristic. Negative values convert to infinite.
double
3.5
<name> .analytic_expansion_max_length
If the length is too far, reject this expansion. This prevents shortcutting of search with its penalty functions far out from the goal itself (e.g. so we don’t reverse half-way across open maps or cut through high cost zones). This should never be smaller than 4-5x the minimum turning radius being used, or planning times will begin to spike.
double
3.0
<name> .reverse_penalty
Heuristic penalty to apply to SE2 node if searching in reverse direction. Only used in allow_reverse_expansion = true.
double
2.0
<name> .change_penalty
Heuristic penalty to apply to SE2 node if changing direction (e.g. left to right) in search.
double
0.05
<name> .non_straight_penalty
Heuristic penalty to apply to SE2 node if searching in non-straight direction.
double
1.05
<name> .cost_penalty
Heuristic penalty to apply to SE2 node for cost at pose. Allows State Lattice to be cost aware.
double
2.0
<name> .rotation_penalty
Penalty to apply for rotations in place, if minimum control set contains in-place rotations. This should always be set sufficiently high to weight against in-place rotations unless strictly necessary for obstacle avoidance or there may be frequent discontinuities in the plan where the plan requests the robot to rotate in place to short-cut an otherwise smooth forward-moving path for marginal path distance savings.
double
5
<name> .retrospective_penalty
Heuristic penalty to apply to SE2 node penalty. Causes State Lattice to prefer later maneuvers before earlier ones along the path. Saves search time since earlier (shorter) branches are not expanded until it is necessary. Must be >= 0.0 and <= 1.0. Must be 0.0 to be fully admissible.
double
0.015
<name> .lattice_filepath
The filepath to the state lattice minimum control set graph, this will default to a 16 bin, 0.5m turning radius control set located in test/ for basic testing and evaluation (opposed to Hybrid-A*’s default of 0.5m).
string
“”
<name> .lookup_table_size
Size of the dubin/reeds-sheep distance window to cache, in meters.
double
20.0
<name> .cache_obstacle_heuristic
Cache the obstacle map dynamic programming distance expansion heuristic between subsiquent replannings of the same goal location. Dramatically speeds up replanning performance (40x) if costmap is largely static.
bool
FALSE
<name> .allow_reverse_expansion
If true, allows the robot to use the primitives to expand in the mirrored opposite direction of the current robot’s orientation (to reverse).
bool
FALSE
<name> .smooth_path
If true, does simple and fast smoothing post-processing to the path from search
bool
TRUE
<name> .smoother.max_iterations
The maximum number of iterations the smoother has to smooth the path, to bound potential computation.
int
1000
<name> .smoother.w_smooth
Weight for smoother to apply to smooth out the data points
double
0.3
<name> .smoother.w_data
Weight for smoother to apply to retain original data information
double
0.2
<name> .smoother.tolerance
Parameter tolerance change amount to terminate smoothing session
double
1.00E-10
<name> .smoother.do_refinement
Performs extra refinement smoothing runs. Essentially, this recursively calls the smoother using the output from the last smoothing cycle to further smooth the path for macro-trends. This typically improves quality especially in the Hybrid-A* planner but can be helpful on the state lattice planner to reduce the “blocky” movements in State Lattice caused by the limited number of headings.
bool
TRUE
<name> .smoother.refinement_num
Number of times to recursively attempt to smooth, must be >= 1.
int
2
နမူနာ
Last updated