human-guided unmanned surface vehicle teamsruk.usc.edu/bio/gupta/ussv.pdfdynamically-feasible,...
TRANSCRIPT
![Page 1: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/1.jpg)
Human-Guided Unmanned Surface Vehicle Teams
Satyandra K. Gupta
Smith International Professor
Aerospace and Mechanical Engineering Department
Computer Science Department
Viterbi School of Engineering
University of Southern California
![Page 2: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/2.jpg)
Pradeep Rajendran
Brual Shah
Shaurya Shriyam
Petr Svec
Main Contributors
![Page 3: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/3.jpg)
Motivation
β’ Many civilian applications can benefit from deploying a human-robot team consisting of several small USVs and one or more human supervisors β Reduces costs
β Improves safety
β Increases operational efficiencies
β’ Representative applications: β Remote/persistent ocean
sensing
β Marine search and rescue
β Maritime operations in congested port environments
β Industrial offshore supply and support Unmanned Surface Vehicle (USV)
![Page 4: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/4.jpg)
Planning Challenges
Need to consider and utilize underlying physics
Need to optimize planning time
Need to proactively manage risk
https://youtu.be/ePDoCPi06rk?t=16s
![Page 5: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/5.jpg)
Research Areas
Risk-aware, COLREGs-
compliant trajectory planning
Planning for target following
with motion prediction
Long distance path planning Energy efficient path planning On-Line Task Decomposition
for Collaborative Surveillance
Dynamics-Aware Reactive
Trajectory Planning
![Page 6: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/6.jpg)
System Architecture
![Page 7: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/7.jpg)
Overview
Develop an optimal long distance path planner that uses A* on visibility graph defined over quadtrees
Develop an path planning algorithm that uses the weather forecast and generate energy efficient paths in time-varying flow fields (e.g. ocean currents)
Generate a meta-model of USVβs dynamics and PID / LQR / backstepping controller
Develop a global, resolution-adaptive, dynamically-feasible, COLREGs-compliant, 5D trajectory planner
Demonstrate capabilities of the planner by performing simulation and physical experiments
Meta-model of
USVβs dynamics
and control
Deliberative
Trajectory
Planner
Dynamically
feasible global
trajectory
Local COLREGs-
compliant
trajectory planner
3 DOF USV
dynamics
simulation model
Dynamically
reachable
motion goal
States estimates of
all vessels using
their intention
models
USVβs low-level
controller
Static obstacles
Path Planner
Start and Goal
Locations Weather Model
![Page 8: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/8.jpg)
Task decomposition for collaborative surveillance by multiple USVs
![Page 9: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/9.jpg)
Deploying a team of robots instead of a single
robot can improve safety, increase operational
efficiencies, and significantly reduce costs.
Collaborative tasks involving one or more team(s)
of robots require spatial partitioning of the region
of interest.
Decomposition of tasks among team(s) of robots
is a very challenging problem because:
β Imperfect information of the environment
β Changing environmental conditions
β Varying performance of the robot as a result
of its interaction with the environment
β Risk of collision with external entities (e.g.,
civilian traffic)
Introduction
![Page 10: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/10.jpg)
A sample frontier based area expansion of a single USV is shown
The cells lying inside region of interest show the values obtained from
velocity map
These values are used to explore the frontier cells in the next time step
But our algorithm only focuses on generating equitable area partitions
Task Decomposition through Frontier- based Exploration
The coverage trajectories arising out
of this area exploration technique is
not outputted as a result
![Page 11: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/11.jpg)
Since all USVs expand their areas, the USV that finishes last is the
bottleneck agent
Time taken by the bottleneck USV is minimized using particle Swarm
Optimization algorithm
The optimization variable is only considered to be the initial placement of
the USVs along the region boundary
Suppose USV π_π finishes in time π‘_ππ. Then we optimize the time π‘_πππππ β taken by bottleneck agent π^β given by
Bottleneck Optimization
![Page 12: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/12.jpg)
Illustrative Examples
![Page 13: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/13.jpg)
Long Distance Path Planner
![Page 14: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/14.jpg)
Topography of a complex environment
USVs are increasingly being used on
long missions with large operating space
and time
The real-world scenarios comprises of
complex polygons and cannot be
represented by standard geometrical
shapes
The available free space in marine
environment changes over time:
- Low tides may make it infeasible to go
through shallow waters
- Environmental restrictions may present
the USVs to go through certain protected
marine regions for certain periods
- Weather induced waves may prohibit
traveling over certain areas due to high
collision risks
Introduction
Quadtree representation of polygon
![Page 15: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/15.jpg)
Related Work
Grid-based methods with constant grid size are inefficient and
computationally expensive. Optimality of the path depends upon the
resolution of the grid
Sampling-based methods produce non-optimal paths and the computation
time is dependent on map specific parameter like ππππ (in RRT*)
Visibility graphs are grid independent and computes optimal paths but
computationally inefficient for online planning
Techniques for Path Planning Representative References
Grid-based approach β’ A*: Hart, Nilsson, and Raphael 1968
β’ Block A*: Yap et. al. 2002
β’ Theta*: Daniel et. al. 2010
Sampling-based approach β’ RRT: LaValle 2001, RRT*: S. Karaman 2010
β’ Roadmaps: Kavraki et al. 1996
β’ Roadmaps and Voronoi diagrams: Bhattacharya
et. al. 2008
Visibility-graph based approach β’ Visibility graphs: Lee et. al. 1978 and Lozano-
Perez et. al. 1979
β’ Visibility binary trees: Rashid et. al. 2013
![Page 16: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/16.jpg)
Overview of Approach
Computation of VG
using Quadtree
Efficient
computation of VG
Heuristic for A*
Search
Focusing the A*
Search
Path Planner
VG: Visibility Graph
![Page 17: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/17.jpg)
Background on Path Planning using Visibility Graph
[1.] Y. H. Liu and S. Arimoto, βPath planning using tangent graph for mobile robots among polygonal and curved obstacles,β Int. J. Robotics Res., vol. 11, no. 4, pp. 376-382, 1992
Represent the polygons imported from the
nautical charts into quadtrees
Options to compute visibility graph:
- Precompute the visibility graph and use Dijkstra's
to compute shortest path
o High computational time (π(π2 log π )) - Compute the visibility graph on the fly and search
using A*
o Computational time can still be high in
challenging scenarios (worst case π(π2 log π ))
The complexity of VG is reduced by only
considering the tangents to the obstacles
without compromising the optimality
The reduced VG is termed as Tangent Graph
(TG)
Quadtree representation of polygon
![Page 18: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/18.jpg)
Computation of Reduced Tangent Graphs using Quadtree Representation
Nodes lying on the
convex hull are
possible
candidates for the
visibility graph
The number of nodes in visibility graphs are
large and makes the search inefficient
Optimal path will not pass through vertices that
lies in the globally concave region
Elimination of globally concave nodes:
- Compute the island regions by combining the
connected component of solid leaf nodes of
quadtree
- Compute the non-intersecting convex hulls for
each island
- Eliminate the nodes that lie in the interior of the
convex hulls
The path is generated by computing the visible
nodes from each expanded node during the
search
The next node is selected based on least-cost
A* fashion
![Page 19: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/19.jpg)
Computational Results
TG: Tangent Graph
RTG: Reduced Tangent Graph
Approximately 70-80% of the nodes from the TG are eliminated
![Page 20: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/20.jpg)
Low computational efficiency of A* on standard tangent graphs is due to:
- Large branching factor of A* search
- Performing collision checks to determine the visible nodes in the graph
The number of collision checks is reduced by eliminating the nodes that lie in the shadow region
Efficient Computation of Visibility Graphs
![Page 21: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/21.jpg)
Method 1 (M1): Uses the tangent graph (TG) with Euclidean distance to goal as heuristic cost (ECU)
Method 2 (M2): Uses the reduced tangent graph (RTG) with Euclidean distance (ECU) and Reduced Collision Checks (RCC)
Method 2 reduces the computation time by approximately 40% over Method 1
Computational Results
![Page 22: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/22.jpg)
Optimal path towards the goal have to
circumvent the largest obstacle that lie on
the straight line path connecting the start
and the goal
The developed new heuristic avoids
expanding nodes on the boundary of the
obstacle and improves the efficiency of the
A* search
Heuristic for A* Search
: Heuristic cost of the path passing right
side of the obstacle
: Heuristic cost of the path passing left
side of the obstacle
![Page 23: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/23.jpg)
Computational Results: New Heuristic
Method 1 (M1): Uses the tangent graph (TG) with Euclidean distance to goal as heuristic cost (ECU)
Method 3 (M3): Uses the reduced tangent graph (RTG) with the developed heuristic (HEU) and Reduced Collision Checks (RCC)
Method 3 reduces the computation time by approximately 70% over Method 1
![Page 24: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/24.jpg)
A* algorithm needs to explore all
possible tangent nodes
Large spatial regions, will have large
number of tangent nodes that needs
to examined
In the scenario shown, the number of
edges on the tangent graph are very
large even the path to goal is easy to
compute
Problem of Large Branching Factor
Pathological scenario where a node in tangent
graph has a large branching factor.
![Page 25: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/25.jpg)
Consider local nodes as candidates for
visibility checks (i.e. nodes that lie within
a circle of predetermined size)
Bias the candidate nodes being
considered towards the goal by
selecting nodes on the convex hull
which intersect the line from the current
state to the destination state
Significantly reduces the candidate
tangent graph nodes to process for
visibility checks Current State
Boundary for
selecting
local nodes
Visible tangent nodes on
convex hull intersecting
line from the current to
the goal node
Visible local
tangent nodes
Focusing A* Search
![Page 26: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/26.jpg)
Method 1 (M1): Uses the tangent graph (TG) with Euclidean distance to goal as heuristic cost (ECU)
Method 4 (M4): Uses the focused search on reduced tangent graph (FS) with the developed heuristic (HEU) and Reduced Collision Checks (RCC)
Method 4 reduces the computation time by approx. 90% over Method 1
Computational performance comes at the cost of marginal loss of optimality and increased path length by approx. 1%
Computational Results: Focused Search
![Page 27: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/27.jpg)
Scaling of our Algorithm
Map Size
(pixels)
Computation Time (in sec)
Theta* Our Approach
1024 x 1024 1.87 0.19
2048 x 2048 16.69 0.22
4096 x 4096 90.784 0.24
Our approach is scalable with the
size of the operating environment
as compared to Theta*
![Page 28: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/28.jpg)
Map size: 100 x 100 km
# of Quadtree nodes: 66425
Depth of the quadtree: 13 i.e.
smallest resolution 12.207 m
# of tangent graph nodes: 2732
# of nodes expanded by focused
search: 711
Computation time by our
algorithm is 12.19 sec
Results: Real Scenario
![Page 29: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/29.jpg)
Assessing Effectiveness of Focused Search
Possibility #1: Next node on the
optimal path is in the local search
region(LSR) πππππππππ‘
Possibility #2: Next node on the
optimal path (NNOP) is NOT in the
local circular search region but lies
on the island that intercepts direct
line to the goal πππ₯π‘πππ‘
Possibility #3: Next node on the
optimal path (NNOP) is NOT in the
local circular search region but
back connection is able to find it
ππππππππ‘
Possibility #4: Next node on the
optimal path cannot be found
(100 β πππππππππ‘
β πππ₯π‘πππ‘
β ππππππππ‘
)
![Page 30: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/30.jpg)
Effect of the size of LSR on πππππππππ‘
Curve increases in super-linear fashion and then begins to saturate as we reach
higher percentage of the mean tangent length.
πππππππππ‘
: Next Node on Optimal Path that lies inside the local search region LSR: Local Search Region
![Page 31: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/31.jpg)
Effect of the size of LSR on πππ₯π‘πππ‘
Plot shows the percentage of remaining nodes on optimal path after eliminating
the nodes that lie inside the local search region i.e. 100 β πππππππππ‘
Approximately 65% of the remaining nodes lie inside the extended search region
πππ₯π‘πππ‘
: Next Node on Optimal Path that lies in side the extended search region LSR: Local Search Region
![Page 32: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/32.jpg)
Effect of the size of LSR on ππππππππ‘
Plot shows the percentage of remaining nodes on optimal path after eliminating the
nodes that lie inside the local and extended search region i.e. 100 β πππππππππ‘
β πππ₯π‘πππ‘
Significant amount of optimal path segment are larger than mean tangents length
πππ₯π‘πππ‘
: Next Node on Optimal Path that lies in side the extended search region LSR: Local Search Region
![Page 33: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/33.jpg)
Effect of the size of LSR on ππππππππ‘
Plot shows percentage of the next
nodes on the optimal path do not lie
in the local and extended search
region and are not discovered by
the back connection approach i.e.
100 β πππππππππ‘
β πππ₯π‘πππ‘
β ππππππππ‘
Selecting the size of LSR
approximately one and half times
the mean tangent length ensures
that the probability of not finding the
optimal path drop below 1%.
For example at mean (i.e. 100%),
the probability of not finding node =
(1-0.5)*(1-0.65)*(1-0.85) = 0.026 =
2.6%
LSR: Local Search Region
![Page 34: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/34.jpg)
Handling Time Varying Free Space
![Page 35: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/35.jpg)
Developed a resolution independent, any-angle path planning algorithm that computed paths on maps of 100 x 100 Km with occupancy of 40% in approximately 2 min (Python implementation)
Computational performance is improved by approx. 90% as compared to tangent graph based approach with marginal loss of optimality
Extended the planner to compute paths in an environment with time varying free space
Summary
![Page 36: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/36.jpg)
Deliberative Trajectory Planner
![Page 37: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/37.jpg)
Introduction
Trajectory planning is a congested and highly dynamic environment with civilian traffic is a non-trivial planning problem
- Need to consider vehicleβs non-linear dynamics and control in planning
- Need to employ a short sense-plan-control cycle especially when operating at high speeds
- Need to follow Coast Guard Collision Regulations (COLREGs)
- Needs to consider intentions of civilian vessels while performing avoidance maneuvers
- Need to find a balanced trade-off among the trajectory length, collision risk, and violation of COLREGs
(source: Map data Β©2013 Google)
A scenario with USV and
civilian vessels moving across
congested harbor scenario
![Page 38: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/38.jpg)
Overview of Approach
Intension
Modelling of
Civilian Vessel
RCAP
Congestion
Metric Generation of
Meta-model
A-RCAP
Trajectory Planner
RCAP: Risk and Contingency Aware Planner
![Page 39: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/39.jpg)
Meta-Model of WAM-V USV14βs Dynamics and Control
WAM-V USV14 simulation model
Control action primitive set
Emergency Control Primitive
![Page 40: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/40.jpg)
Intention Modelling of Civilian Vessels
t = 5s
t = 8s
t = 13s
t = 22s
t = 32s
Dynamic
obstacle
USV14 Estimated
current state of
civilian vessel
Static obstacle
regions
Physical
characteristic of
civilian vessel
Classifier Motion
prediction
model
Predicted state
of civilian vessel
Uncertainty in Civilian Vessels: - Mean: Lies on trajectory provided by intention model
- Variance:
where c = cos (πππ) & s = π ππ(πππ)
![Page 41: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/41.jpg)
Risk and Contingency- Aware Trajectory Planner (RCAP)
Developed a 5D π₯, π¦, π, π’, π‘ π lattice-based global planner RCAP
Incorporated the reciprocal behavior of civilian vessels into planning
Reduced the collision rate by integrating the contingency maneuvers directly into the search
![Page 42: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/42.jpg)
42
A set of control action primitive
divided into three regions
Adaptive Time Scaling of
Control Primitives in RCAP (A-RCAP)
Adaptively scaled the control primitives to speed-up the search
![Page 43: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/43.jpg)
Congestion Metric
![Page 44: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/44.jpg)
Simulation Results
61-87% reduction in collision rate as compared to COLREGs-compliant VO-based planner
Reduced the distance travelled by additional ~4% by computing globally optimal trajectories
![Page 46: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/46.jpg)
Energy-efficient Path Planner
![Page 47: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/47.jpg)
Introduction
Performance of the USVs are influenced by
the underlying medium flow in which they
operate
Usually, USVs follow a shortest distance
path by using a feedback controller to
compensate for the disturbances, but that
leads to: - Significant consumption of energy
- Wear and tear of actuator over a long period
of operation
It will be advantageous for the USVs to
exploit the medium flows instead of
overcoming it, that provides: - Increase in overall mission duration
![Page 48: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/48.jpg)
Overview of Approach
Motion Model Start Time
Optimization
Admissible
Heuristic for
A* Search
Path Planner
![Page 49: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/49.jpg)
Motion Primitives & Motion Model
The motion primitive set consists of actuation
action and a special free-flow action
Transition from the current state π to
neighboring state π β² is decided by velocity of
the medium flow and thrust-producing action
π£π ,π‘π = Current velocity of the robot (or USV)
β π£π ,π‘π,π
= Component along orthogonal direction
β π£π ,π‘π,π·
= Component along desired direction
π£π ,π‘π = Current velocity of the medium
β |π£π ,π‘π |sin(ππ) = Component along orthogonal dir.
β |π£π ,π‘π |cos(ππ) = Component along desired dir.
Forward Velocity
|π£π ,π‘π| = π£π ,π‘
π,π· + |π£π ,π‘π |cos (ππ)
Assumption: Low level controllers of the
vehicle are able to maintain the desired dir.
![Page 50: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/50.jpg)
Admissible Heuristic #2
π‘πππ’ππ
![Page 51: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/51.jpg)
Admissible Heuristic #3
Heuristic #2 does not account for the orthogonal deviation from the desired path to reach the goal
Free-flow actions have to compensate for its deviation from the desired direction to reach the goal
New cost incurred per unit length
advancement towards the goal for free-flowing action is given by:
![Page 52: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/52.jpg)
Simulation Results
Cost of actuation action πΆππ‘ = 6 per min
Cost of free-flowing action πΆππ‘ = 1.2 per min
Discrete time interval πΏπ‘ = 10 min
Experimental Scenarios A: β Medium flow with constant magnitude of 6 m/s o Constant flow direction at 30π and 170π
o Rotating Medium flow at rate of 0.1π per min with initial direction set at 3300
o Rotating Medium flow at rate of 0.2π per min with initial direction set at 3100
![Page 53: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/53.jpg)
Simulation Results (cont.)
Experimental Scenarios B: β Medium flow with randomly generate magnitude profile o Constant flow direction at 30π and 90π
o Randomly generate direction profile with maximum change of 5π in each discrete time
interval πΏπ‘ = 10 min
o Rotating Medium flow at rate of 0.1π per min with initial direction set at 3300
o Rotating Medium flow at rate of 0.2π per min with initial direction set at 3100
The developed admissible heuristic improves the computational performance of the planner by approximately 90% as compared to baseline heuristic
![Page 54: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/54.jpg)
Start Time Optimization
Weather forecast can be used to determine the optimal time when the vehicle should start the mission π‘π π‘πππ‘
Deliberative lattice-based A* planner compute optimal cost (πΆβ) from start to goal state at each sampled time π‘π π‘πππ‘
The optimizer adaptively samples the time and determines the resolution-optimal start time
Optimizer Deliberative
Planner
π‘π π‘πππ‘
πΆβ
π‘π π‘πππ‘β
![Page 55: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/55.jpg)
Simulation Results
B C D
A.1 A.2 A
![Page 56: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/56.jpg)
Planning for Target Following with Motion Prediction
![Page 57: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/57.jpg)
Safe and efficient following of time
varying motion goal π₯πΊ = π₯, π¦, π, π£ π
in environment with obstacles
presents multiple challenges
β USV following capability is influenced by
its dynamics π₯ = ππ π₯, π’ , motion
characteristics of motion goal, and
distribution of obstacles
β USV does not know trajectory of target
in advance
β USV needs to keep motion goal within
specified distance range (ππππ , ππππ₯)
β USV needs to maintain sufficient
velocity and have capability to negotiate
sharp turns
Introduction
![Page 58: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/58.jpg)
Approach
Dense action set Sparse action set
Trajectory Planner: Multi-resolution
Control Action Set
Motion Goal Prediction
![Page 60: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/60.jpg)
Reactive Trajectory Planner
![Page 61: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/61.jpg)
Local, Adaptive Sampling-Based Trajectory Planner
Developed generalized Velocity
Obstacles based planner that
computes a dynamically reachable,
COLREGs-compliant motion goal o Generalizes state-of-the-art OA
approaches [Wilkie, D. et al., 2009;
Berg et al., 2012; Bareiss et al., 2013]
to any non-linear, non-holonomic
dynamic model and controller
o Determines motion goals reachable along
collision-free state space trajectories
through resolution adaptive sampling
o Can handle constraints on lower-level
control input to ensure COLREGs
Expanded set of motion goals in classical, uniform sampling-based local planning
Expanded set of motion goals in adaptive, non-uniform sampling-based local planning
![Page 62: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/62.jpg)
Local Trajectory Planner: Adaptive Sampling Based Search
USV
Target
Sampled collision-free motion goals
Sampled collision motion goals
Selected motion goal
Civilian vessel
USV
Selected motion goal
Sampled collision-free motion goals
Target
Expanded set of motion goals in classical, uniform sampling-based
local planning
Expanded set of motion goals in adaptive, non-uniform sampling-
based local planning
![Page 63: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/63.jpg)
Adaptive sampling based on temporal-spatial distribution of obstacle vessels and history of detected collisions during the search o Control-imposed obstacle
regions adaptively expanded
Two types of sampling: 1) Prioritized sampling of entire
set of state space trajectories in each sampling cycle
2) Sampling of individual state space trajectories for collision detection
Local Trajectory Planner: Adaptive Sampling Based Search (cont.)
![Page 64: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/64.jpg)
Cost computation of sampled motion goals π π±π = ππ‘π + 1 β π π‘π + ππππ,π πππ‘ + πΒ¬πΆππΏπ πΈπΊπ o π‘π is the true time to reach the motion goal
o π‘π is heuristic estimate of the time to reach goal state
oπ controls the breadth of the sampling-based search
o ππππ,π πππ‘ is the cost for entering the soft obstacle region
Min. distance at the closest point of approach
Collision penalty for entering soft obstacle region
Distance threshold for entering soft collision region
Local Trajectory Planner: Adaptive Sampling Based Search (cont.)
![Page 65: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/65.jpg)
Simulation Setup
1000 randomly generated
evaluation scenarios
o 200 x 200 m test scene
o 3 DOF model of USV
dynamics
o USVβs max. speed is 2.5 m/s
o 20 civilian, randomly
distributed vessels with
varying speed up to 3 m/s
o following COLREGs was not
considered
Intel(R) Core(TM) i7-2600
CPU @ 3.4 GHz computer
with 8GB RAM
USV
Goal
![Page 66: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/66.jpg)
Simulation Results (Cont.)
Comparison of collision rate and search time of the full
sampling-based OA planner to Velocity Obstacles
[Fiorini and Shiller, 1998]
1) Planner considers vehicleβs dynamics and control in the search for a motion goal
2) However, its collision rate is higher than that of the Velocity Obstacles!
3) Planning is ~2.5 times slower than that of the Velocity Obstacles
4) Time to reach the goal is comparable to the time taken by Velocity Obstacles
![Page 67: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/67.jpg)
Simulation Results (Cont.)
Comparison of collision rate and search time of the full and
adaptive sampling-based OA planners to Velocity Obstacles
[Fiorini and Shiller, 1998]
1) Planners consider vehicleβs dynamics and control in the search for a motion goal
2) The search speed of the adaptive sampling-based planner comparable to
the speed of the Velocity Obstacles
3) This results in more than 50% reduction in the collision rate
4) Time to reach the goal is comparable to the time taken by Velocity Obstacles
![Page 68: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/68.jpg)
Wave-Aware Trajectory Planning for Unmanned Surface Vehicles Operating in Congested Environments
(Work done in collaboration Florida Atlantic University)
![Page 69: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/69.jpg)
Motivation
Port policing
Search and rescue
Port assistance
60 %
46 %
70 %
Causes
Failure Modes
Failure Locations
Annual Report on Marine Casualties
These are opportunities for autonomous operations !
![Page 70: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/70.jpg)
Autonomous Operations
β’ Automated trajectory planning is a fundamental requirement
Trajectory Tracking Control
Mission specification Survey region XYZ and report findings to base station.
Safely move from point A to point B as fast as possible.
Set throttle and steering rate profile
Task specification
Trajectory Planning
Depth
check
Current
check
Traffic
check Courtesy: Google Maps
Decompose mission Into tasks
![Page 71: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/71.jpg)
Traditional Trajectory Planning
Safely move from point A to point B as fast as possible.
Reactive Dynamic Obstacle Avoidance
Trajectory Planning
Nominal Trajectory Planning
![Page 72: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/72.jpg)
Traditional Trajectory Planning
Nominal trajectory planning
Reactive collision avoidance
Human expert behavior
Time taken : 446 s Time taken : 364 s
+
![Page 73: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/73.jpg)
Our Idea
β’ Deliberative planning for handling β Large, uncertain dynamic obstacles β Robot motion uncertainty β Risk of traversing waves
![Page 74: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/74.jpg)
Challenges
β’ NaΓ―ve A* implementation is slow β Large, permeable dynamic obstacles
β Large state-space
β’ Handling uncertainty using Monte-Carlo sampling is slow
...
...
State lattice (π₯, π¦, π, π‘)
![Page 75: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/75.jpg)
Approach Outline
Precompute heuristics
Off
line
Characterize wavefields
of traffic vessels
Characterize trajectory controller tracking error
statistics
β’ Monte-Carlo simulations under various β sea-states
β wave amplitudes
β wave directions
β’ Transition trajectories using GPUs [Thakur2012]
![Page 76: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/76.jpg)
Approach Outline
Precompute heuristics
Off
line
Characterize wavefields
of traffic vessels
Characterize trajectory controller tracking error
statistics
π
Wave exposure safety model
![Page 77: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/77.jpg)
Cost Function
Travel time
Failure penalty
COLREGs penalty
![Page 78: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/78.jpg)
Approach Outline
Precompute heuristics
Off
line
Characterize wavefields
of traffic vessels
Characterize trajectory controller tracking error
statistics
Dynamic obstacle heuristics Static obstacle heuristics
β’ Heuristics
β’ Adaptive motion goals
β Use proximity to static and dynamic obstacles to locally increase lattice resolution
![Page 79: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/79.jpg)
Results: Physical Experiments
Faster
Our method Alternative method
![Page 80: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/80.jpg)
Results: Simulation Experiments
Safer
Faster
Alternative methods Our method
Improved computational efficiency
![Page 81: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/81.jpg)
Results: Effect of Different MTBF
Safe Conservative Behavior Safe Aggressive Behavior
Setting waves to be more dangerous
Setting waves to be less dangerous
![Page 82: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/82.jpg)
Results: Effect of Increasing Perception Uncertainty
Safe Aggressive Behavior Safe Conservative Behavior
Low Perception Uncertainty High Perception Uncertainty
![Page 83: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/83.jpg)
Results: Impact of Increasing Obstacle Count
Stress-Test Map
![Page 84: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/84.jpg)
β’ Demonstrated the feasibility of trajectory using deliberative planning in a few seconds while
β Handling large, uncertain dynamic obstacles β Handling motion uncertainty β Assessing the risk of traversing semi-permeable
dynamic obstacles
Conclusion
![Page 85: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/85.jpg)
B. Shah and S.K. Gupta. Long distance path planning for unmanned surface vehicles in complex marine environment. IEEE Journal of
Oceanic Engineering, Accepted for publication.
S. Shriyam. B. C. Shah, and S.K. Gupta. Decomposition of collaborative surveillance tasks for execution in marine environments by a
team of unmanned surface vehicles. ASME Journal of Mechanisms and Robotics, 10(2):025007-025007-7, 2018.
P. Rajendran, T. Moscicki, J. Wampler, B. C. Shah, K. D. von Ellenrieder, and S. K. Gupta. Wave-aware trajectory planning for
unmanned surface vehicles operating in congested environments. IEEE International Symposium on Safety, Security, and Rescue
Robotics, Philadelphia, PA, USA, August 2018.
S. Shriyam, and S.K. Gupta. Incorporating potential contingency tasks in multi-robot mission planning. IEEE International Conference
on Robotics and Automation, Brisbane, Australia, May 2018.
S. Shriyam, B.C. Shah, and S.K. Gupta. On-line task decomposition for collaborative surveillance of marine environment by a team of
unmanned surface vehicles. ASME Mechanisms and Robotics Conference, Cleveland, OH, August 2017.
B. C. Shah, P. Svec, I.R. Bertaska, W. Klinger, A.J. Sinisterra, K. v. Ellenrieder, M. Dhanak, and S.K. Gupta. Resolution-adaptive risk-
aware trajectory planning for surface vehicles operating in congested civilian traffic. Autonomous Robots, 40(7): 1139β1163, 2016.
B.C. Shah and S.K. Gupta. Speeding up A* search on visibility graphs defined over quadtrees to enable long distance path planning
for unmanned surface vehicles. International Conference on Automated Planning and Scheduling, London, UK, June 2016.
I.R. Bertaska, B. Shah, K. von Ellenrieder, P. Svec, W. Klinger, A.J. Sinisterra, M. Dhanak, and S.K. Gupta. Experimental evaluation of
automatically-generated behaviors for USV operations. Ocean Engineering, 106:496β514, 2015.
P. Svec, A. Thakur, E. Raboin, B.C. Shah, and S.K. Gupta. Target following with motion prediction for unmanned surface vehicle
operating in cluttered environments. Autonomous Robots, 36(4): 383-405, 2014.
B. C. Shah, P. Svec, I. R. Bertaska, W. Klinger, A. J. Sinisterra, K. von Ellenrieder, M. Dhanak, and S.K. Gupta. Trajectory planning
with adaptive control primitives for autonomous surface vehicles operating in congested civilian traffic. IEEE/RSJ International
Conference on Intelligent Robots and Systems, Chicago, IL, USA, September 14-18, 2014.
R. B. Bertaska, J. Alvarez, A. J. Sinisterra, K. Ellenrieder, M. Dhanak, B. Shah, P. Svec, and S.K. Gupta. Experimental evaluation of
approach behavior for autonomous surface vehicles. ASME Dynamic Systems and Control Conference, Stanford University, Palo Alto,
October 2013.
References
![Page 86: Human-Guided Unmanned Surface Vehicle Teamsruk.usc.edu/bio/gupta/USSV.pdfdynamically-feasible, COLREGs-compliant, 5D trajectory planner Demonstrate capabilities of the planner by performing](https://reader030.vdocument.in/reader030/viewer/2022040302/5e7db5ae80250650ca60ecda/html5/thumbnails/86.jpg)
For more information please contact: Satyandra K. Gupta ([email protected])