Robot Manipulation Planning with Co-safe LTL Goals · Co-safe LTL formula over these atomic...
Transcript of Robot Manipulation Planning with Co-safe LTL Goals · Co-safe LTL formula over these atomic...
![Page 1: Robot Manipulation Planning with Co-safe LTL Goals · Co-safe LTL formula over these atomic propositions. 4 Temporal Manipulation Tasks • State of the art in manipulation – A](https://reader033.fdocuments.net/reader033/viewer/2022042612/5f4fb8fcdf27e54bc0072d5e/html5/thumbnails/1.jpg)
1
Robot Manipulation Planning with Co-safe LTL Goals
Joint work with Morteza Lahijanian, Lydia E. Kavraki, Moshe Y. Vardi
Keliang He Rice University
ExCAPE Meeting 2015
![Page 2: Robot Manipulation Planning with Co-safe LTL Goals · Co-safe LTL formula over these atomic propositions. 4 Temporal Manipulation Tasks • State of the art in manipulation – A](https://reader033.fdocuments.net/reader033/viewer/2022042612/5f4fb8fcdf27e54bc0072d5e/html5/thumbnails/2.jpg)
2
Robot Manipulation
Packing/unpacking
Preparing food
Serving at a Cafe
![Page 3: Robot Manipulation Planning with Co-safe LTL Goals · Co-safe LTL formula over these atomic propositions. 4 Temporal Manipulation Tasks • State of the art in manipulation – A](https://reader033.fdocuments.net/reader033/viewer/2022042612/5f4fb8fcdf27e54bc0072d5e/html5/thumbnails/3.jpg)
3
Task Formulation
Objects
Locations of interest with labels
} Object X is at a location with label Y
Atomic Propositions
Manipulation Task: Co-safe LTL formula over these atomic propositions
![Page 4: Robot Manipulation Planning with Co-safe LTL Goals · Co-safe LTL formula over these atomic propositions. 4 Temporal Manipulation Tasks • State of the art in manipulation – A](https://reader033.fdocuments.net/reader033/viewer/2022042612/5f4fb8fcdf27e54bc0072d5e/html5/thumbnails/4.jpg)
4
Temporal Manipulation Tasks
• State of the art in manipulation– A to B (Reachability)
• Complex tasks involve many steps– Detailed order may be irrelevant to user– Choice may determine feasibility
Offer snacks to all guests and ask for tip from the guests already served.
![Page 5: Robot Manipulation Planning with Co-safe LTL Goals · Co-safe LTL formula over these atomic propositions. 4 Temporal Manipulation Tasks • State of the art in manipulation – A](https://reader033.fdocuments.net/reader033/viewer/2022042612/5f4fb8fcdf27e54bc0072d5e/html5/thumbnails/5.jpg)
5
Challenges
≥6 DoF Manipulator Many ObjectsComplex Workspace
Temporal Taskcontinuous search space
Smaller Problem Smaller Problem Smaller Problem...? Abstraction
? Solving Strategy
HUGE
![Page 6: Robot Manipulation Planning with Co-safe LTL Goals · Co-safe LTL formula over these atomic propositions. 4 Temporal Manipulation Tasks • State of the art in manipulation – A](https://reader033.fdocuments.net/reader033/viewer/2022042612/5f4fb8fcdf27e54bc0072d5e/html5/thumbnails/6.jpg)
6
Planning Framework
C-SpaceDecomposition
3-Layered Synergistic Planner
(Bhatia '11)
ContinuousTrajectory
Abstraction
Modified 3-Layered Synergistic Planner
DFA
Complexity: Dynamics of the systemNavigation ProblemManipulation Problem
Complexity: Dimensionality of manipulation
Objects WorkspaceActions
Planning Task(Co-safe LTL)
[ICRA 2015]
![Page 7: Robot Manipulation Planning with Co-safe LTL Goals · Co-safe LTL formula over these atomic propositions. 4 Temporal Manipulation Tasks • State of the art in manipulation – A](https://reader033.fdocuments.net/reader033/viewer/2022042612/5f4fb8fcdf27e54bc0072d5e/html5/thumbnails/7.jpg)
7
Abstraction - Overview
•Labeled, weighted graph
![Page 8: Robot Manipulation Planning with Co-safe LTL Goals · Co-safe LTL formula over these atomic propositions. 4 Temporal Manipulation Tasks • State of the art in manipulation – A](https://reader033.fdocuments.net/reader033/viewer/2022042612/5f4fb8fcdf27e54bc0072d5e/html5/thumbnails/8.jpg)
8
Abstraction - Node
• Composition of– Action being performed– Object locations– End effector location– Object in gripper
Action
Object 1 Location
Object 2 Location
Robot Location
Object in Gripper
Grasp
Loc 2
Loc 4
Loc 2
Empty
An abstraction graph node
![Page 9: Robot Manipulation Planning with Co-safe LTL Goals · Co-safe LTL formula over these atomic propositions. 4 Temporal Manipulation Tasks • State of the art in manipulation – A](https://reader033.fdocuments.net/reader033/viewer/2022042612/5f4fb8fcdf27e54bc0072d5e/html5/thumbnails/9.jpg)
9
Abstraction - Action Graph• Currently 4 actions
– Grasp, Place • Precomputed primitive• Performed using visual
feedback– Hold, Move
• Require planning• Must consider model of
environment
• Could be extended to add more actions
Action Graph
![Page 10: Robot Manipulation Planning with Co-safe LTL Goals · Co-safe LTL formula over these atomic propositions. 4 Temporal Manipulation Tasks • State of the art in manipulation – A](https://reader033.fdocuments.net/reader033/viewer/2022042612/5f4fb8fcdf27e54bc0072d5e/html5/thumbnails/10.jpg)
10
Abstraction - Example
![Page 11: Robot Manipulation Planning with Co-safe LTL Goals · Co-safe LTL formula over these atomic propositions. 4 Temporal Manipulation Tasks • State of the art in manipulation – A](https://reader033.fdocuments.net/reader033/viewer/2022042612/5f4fb8fcdf27e54bc0072d5e/html5/thumbnails/11.jpg)
11
Synergistic Layers
Only some actions require motion planning
Do not reconsider actions already
foundTask
![Page 12: Robot Manipulation Planning with Co-safe LTL Goals · Co-safe LTL formula over these atomic propositions. 4 Temporal Manipulation Tasks • State of the art in manipulation – A](https://reader033.fdocuments.net/reader033/viewer/2022042612/5f4fb8fcdf27e54bc0072d5e/html5/thumbnails/12.jpg)
12
Example - Baxter Cafe
First trash the empty can, then offer snacks to all guests and ask for tip from the guests already served.
Planner picks one of many executions
Never undo an action
![Page 13: Robot Manipulation Planning with Co-safe LTL Goals · Co-safe LTL formula over these atomic propositions. 4 Temporal Manipulation Tasks • State of the art in manipulation – A](https://reader033.fdocuments.net/reader033/viewer/2022042612/5f4fb8fcdf27e54bc0072d5e/html5/thumbnails/13.jpg)
13
Example - PR2 Simulation
Object of interest
Region of interest
Task: Move object of interst to region of interest
Intelligently remove objects in the way
![Page 14: Robot Manipulation Planning with Co-safe LTL Goals · Co-safe LTL formula over these atomic propositions. 4 Temporal Manipulation Tasks • State of the art in manipulation – A](https://reader033.fdocuments.net/reader033/viewer/2022042612/5f4fb8fcdf27e54bc0072d5e/html5/thumbnails/14.jpg)
14
Runtime
Scenario # Objects # DFA States# Reachable
Nodes in Product
Avg Total Task Planning
Time (s)
Avg Total Motion Planning
Time (s)
Baxter 3 10 19,370 0.94 0.70
PR2 4 2 44,100 2.76 12.32
PR2 4 8 75,511 4.48 8.07
PR2 4 27 498,000 33.12 31.15
All results averaged over 50 runs
![Page 15: Robot Manipulation Planning with Co-safe LTL Goals · Co-safe LTL formula over these atomic propositions. 4 Temporal Manipulation Tasks • State of the art in manipulation – A](https://reader033.fdocuments.net/reader033/viewer/2022042612/5f4fb8fcdf27e54bc0072d5e/html5/thumbnails/15.jpg)
15
Take-aways
• Formulate manipulation tasks in co-safe LTL– Challenge: high dimensional continuous space
• First work to address this problem– Through novel abstraction and synergistic planning
• Possible future directions:– Faster discrete search– Apply abstraction to other manipulation task