![Page 1: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/1.jpg)
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
Algorithms for Sensor-Based Robotics:Sampling-Based Motion Planning
Computer Science 336http://www.cs.jhu.edu/~hager/Teaching/cs336
Professor Hagerhttp://www.cs.jhu.edu/~hager
![Page 2: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/2.jpg)
Recall Earlier Methods
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 3: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/3.jpg)
16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato
The Basic Idea
• Capture the connectivity of Qfree by a graph or network of paths.
![Page 4: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/4.jpg)
16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato
RoadMap Definition
• A roadmap, RM, is a set of trajectories (i.e. f ( t, qA, qB ) ) such that for all qstart ∈ Qfree and qgoal ∈ Qfree can be connected by a path:
• The three ingredients of a roadmap– Accessibility: There is a path from qstart∈ Qfree to some q’ ∈ RM
– Departability: There is a path from some q’’∈ RM to qgoal ∈ Qfree
– Connectivity: there exists a path in RM between q’ and q’’
![Page 5: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/5.jpg)
16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato
RoadMap Path Planning
1. Build the roadmapa) nodes are points in 𝑄𝑄free or its
boundaryb) two nodes are connected by an
edge if there is a free path between them (i.e. f ( t, qA, qB ))
2. Connect qstart and qgoal points to the road map at point q’ and q’’, respectively
3. Find a path on the roadmap between q’ and q’’. The result is a path in Qfree from start to goal
![Page 6: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/6.jpg)
Roadmap
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 7: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/7.jpg)
Roadmap
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 8: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/8.jpg)
A Hard Problem
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 9: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/9.jpg)
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
The Way Forward
• The hard part of path planning is to somehow represent the free configuration space
– Recall the basic PSPACE-hard results of Reif and Canny.– Recall the limitations of potential fields and even navigation functions.
• Probabilistic RoadMap Planning (PRM) by Kavraki, 1996– samples to find free configurations– connects the configurations (creates a graph)– is designed to be a multi-query planner
• Expansive-Spaces Tree planner (EST) and Rapidly-exploring Random Tree planner (RRT)
– are appropriate for single query problems
![Page 10: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/10.jpg)
Remember the Basic Problem
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 11: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/11.jpg)
How about Rolling the Dice?
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 12: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/12.jpg)
Sample
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 13: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/13.jpg)
Discard
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 14: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/14.jpg)
Connect
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 15: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/15.jpg)
Discard #2
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 16: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/16.jpg)
Finally, a Roadmap
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 17: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/17.jpg)
The Complete Algorithm
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 18: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/18.jpg)
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
Why Are Probabilistic Methods Effective?• Mainly, they do not try to construct the free configuration space
– depend on local sampling of configurations (depends on efficient collision detection)
– make use of relatively “dumb” planners to connect nodes
• They are often probabilistically complete– the probability of correct
solution can be ashigh as desired with enough time
![Page 19: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/19.jpg)
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
Some Minor Issues
• Ideally, the resulting graph will be connected– if it is not it might mean the space is disconnected– it might mean we didn’t try hard enough (what is a hard case?)
• We haven’t specified what the distance or path planner is
• To use for queries we must– connect the start and goal configurations to the roadmap (usually just
treat like nodes and perform same algorithm)– perform a graph search on the resulting graph– if desired, smooth the resulting path a bit
• Interesting to note that the queries can be used to add more nodes to the graph
![Page 20: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/20.jpg)
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
Sampling and Neighbors
• Uniform sampling of configurations– need to take care that rotations are “fairly” sampled
• Selecting closest neighbors: kd-tree– Given: a set S of n points in d-dimensional space– Recursively
• choose a plane P that splits S about evenly (usually in a coordinate dimension)• store P at node• apply to children Sl and Sr
– Requires O(dn) storage, built in O(dn log n) time– Query takes O(n1-1/d + m) time where m is # of neighbors
• asymptotically linear in n and m with large d
• Selecting closest neighbors: cell-based method– when each point is generated, hash to a cell location
![Page 21: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/21.jpg)
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
Distance Functions
• Distance between two configurations should reflect the likelihood that the planner will fail to find a path– close points, likely to succeed– far away, less likely
• Ideally, this is probably related to the area swept out by the robot– very hard to compute exactly– usually heuristic distance is used
• Typical approaches– Euclidean distance on some embedding of c-space– Alternative is to create a weighted combination of translation and
rotational “distances”
– Efficiency varies greatly depending on embedding
![Page 22: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/22.jpg)
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
Local Planner
• The local planner should be reasonably fast and can be simple to implement
• A simple way is to do subdivision on straight line path– Decompose motion into a straight lines in configuration space– Split the line in half; check for collision– If none, recurse on halves until distance is small
![Page 23: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/23.jpg)
PRM Applied to a Point Robot
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 24: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/24.jpg)
PRM for a Serial Chain
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 25: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/25.jpg)
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
Sampling Strategies
• The narrow corridor problem– probability of finding a path related to joint visibility area under
uniform sampling
• Other approaches:– Bridge planner
• sample randomly• check pairs in collision to see if midpoint (or random distance) is not• Somehow use generalized Voronoi diagrams
– Visibility-based sampling• only keep configuration that
– cannot be connected to an existing component, or– connect 2 existing components
![Page 26: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/26.jpg)
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
Sampling Strategies
• Recall the narrow corridor problem– probability of finding a path related to joint visibility area under
uniform sampling– Few samples will be available in here
• Other approaches:– sampling near obstacles
• Obstacle-Based PRM (OBPRM)• find samples in obstacles (uniform)• pick a random direction v• search for a free configuration in direction v
– Gaussian OBPRM• generate a random sample• generate a Gaussian sample around the sample• only keep the sample if the Gaussian sample is in collision
– Dilated obstacles• allow some interpenetration to enhance the likelihood of finding paths,
then “fix up” later.
![Page 27: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/27.jpg)
Obstacle Sampling
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 28: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/28.jpg)
Gaussian Sampling PRM
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 29: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/29.jpg)
Bridge Sampling
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 30: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/30.jpg)
Visibility Sampling
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 31: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/31.jpg)
Importance Sampling
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 32: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/32.jpg)
Avoiding Cycles
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 33: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/33.jpg)
What about Single Queries?
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 34: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/34.jpg)
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
Single Query Planners
• Use ESTs (expansive space trees) and RRTs (rapidly exploring random trees.
– also work for nonholonomic and kinodynamic planning– each of these is initial condition dependent (requires integration) and thus it is
hard to build a graph offline
• Generally build two trees from start and goal configurations– each tree grows toward the other– once linkage is made defines a path from start to goal
• Key is to define sampling strategies that focus on– unexplored areas of the free space– move toward start or goal
• For completeness, necessary to show that the algorithm will eventually cover the entire space
![Page 35: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/35.jpg)
Tree-Based Motion Planning
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 36: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/36.jpg)
Expansive Space Trees (EST)
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 37: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/37.jpg)
BiDirectional Motion Planning
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 38: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/38.jpg)
Rapidly-exploring Random Trees
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 39: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/39.jpg)
Improvements
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 40: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/40.jpg)
Improvements
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 41: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/41.jpg)
An Important Point
• Note that ESTs (also RRTs with modifications) can model non-holonomic and complex systems.
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
Sampling:1) Choose configuration variables (e.g
steering and acceleration) at random.
2) Integrate kinematics forwardto get a path segment
3) Check the path for collision
![Page 42: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/42.jpg)
Sampling-base Roadmap of Trees
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 43: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/43.jpg)
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
Analysis of PRM• Goal: show probabilistic completeness:
– Suppose that a,b ∈ Qfree can be connected by a free path. PRM is probabilistically complete if, for any (a,b)
limn→∞ Pr[(a,b)FAILURE] = 0
where n is the number of samples used to construct the roadmap• Basic idea:
– reduce the path to a set of open balls in free space– figure out how many samples it will take to generate a pair of points
in those balls– connect those points to create a path
• Assuming that a path between a and b exists, the probability of that PRM will fail depends on1. The length of the path2. The distance of the path to the obstacle3. The number of configuration in the roadmap
![Page 44: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/44.jpg)
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
Analysis of PRM
• A path from a to b can be described by a function γ: [0,1] → Qfreewith 𝛾𝛾 0 = 𝑎𝑎 and 𝛾𝛾 1 = 𝑏𝑏
• Let clr(γ) be the minimum distance between γ and any obstacle.
• Let µ be a volume measure on the space, for set S ∈ ℝ𝑑𝑑 , 𝜇𝜇(𝑆𝑆) is the volume of set S
• Bρ(x) is a ball centered at x of radius ρ
• For uniform sampling of A ⊂ Qfree the probability that a random point 𝑥𝑥 ∈ 𝑄𝑄free, then Pr(x ∈ A) = µ(A)/µ(Qfree)
![Page 45: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/45.jpg)
A Simple Idea
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
![Page 46: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/46.jpg)
Completeness
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
𝜌𝜌
![Page 47: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/47.jpg)
The Proof Sketch
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
n
![Page 48: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/48.jpg)
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
Understanding Complexity
• Relationship between “difficulty” of space and number of samples (avoids dependence on length of path which is unknown)
• Relies on so-called (ε, α, β) expansiveness of space– ε measures the fraction of space reachable from any point
• called ε-good if all points “see” at least ε of the space• ∀𝑥𝑥 ∈ 𝑄𝑄𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓 𝜇𝜇(𝑟𝑟𝑟𝑟𝑎𝑎𝑟𝑟𝑟(𝑥𝑥)) ≥ 𝜖𝜖 𝜇𝜇(𝑄𝑄𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓)
– lookoutβ of a set S is the set of points that can see at least β of Qfree - S
• A space is (ε α β) expansive if– it is ε-good– for any connected S⊆ Qfree µ(lookoutβ(S)) ≥ α µ(S)– intuitively, this capture how easy it is to choose points that link up the space
![Page 49: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/49.jpg)
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
Expansive Spaces
x
Reachability
![Page 50: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/50.jpg)
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
Expansive Spaces
Only a small subset of A can see a large fraction of points in B
A B
![Page 51: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/51.jpg)
An Example
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
W WW
w
Given n random samples, what are the odds the resulting graph is connected?
W
![Page 52: Algorithms for Sensor-Based Robotics: Sampling …sleonard/week06.pdf16-735, Howie Choset, with significant copying from G.D. Hager who loosely based his notes on notes by Nancy Amato](https://reader033.vdocuments.us/reader033/viewer/2022050406/5f8328fbd518423e8a24c138/html5/thumbnails/52.jpg)
10/8/2019 CS 336, G.D. Hager (w/slides from S. LaValle, E Plaku)
Summary
• PRS --- a basic sampling-based planner– useful for multiple queries– can be made efficient using various sampling and evaluation tricks
• EST and RRT– single query planners– generate trees and merge them– again many variations based on applications
• Analysis– general idea of completeness– particular proof based on sampling– other ideas based on structure of environment
• Other Applications