CS 4803/8803: RIP - Robot Intelligence - Planning in Action
TR 1:35-2:55, Bunger-Henry 380
Fall 2009

Instructor: Mike Stilman

Office hours:
CCB 254 Thursday 12:00 - 1:00 or by appointment.

TA: Jon Scholz
jkscholz@gatech.edu
Office hours:
By email or appointment.

Tentative Syllabus

Fall 2008 Class Projects

For Discussion, Group Building and Collaboration on Projects: Course Wiki

Summary:
We discuss algorithms for robots and other complex systems that make intelligent decisions in high dimensional or continuous spaces of options. Intelligent decisions take into account both present and future constraints on the system. The course will cover methods for planning with symbolic, numerical, geometric and physical constraints. Topics will range from classical and stochastic planning to continuous robot domains and hybrid control of dynamic systems.

Schedule:
(PDFs of Slides will be posted During course):

Classical Planning

Aug. 18: Introduction, Logic, Situation Calculus: [PDF]
Aug. 20: Advantages of SitCalc, STRIPS Operators, First Planner: [PDF]
- Ch. 21-22.1 "Artificial Intelligence: A New Synthesis," Nilsson
- Ch. 7.1-7.5, 8.1-9.3, 10.1-10.3 "Artificial Intelligence: A Modern Approach," Russel, Norvig
- H. Enderton, "A Mathematical Introduction to Logic," (If you are really curious)
Aug. 25: Shakey, Properties of Planners, State Space vs. Plan Space: [PDF]
- Ch. 22.2 "Artificial Intelligence: A New Synthesis," Nilsson
- Ch. 11.1-11.3 "Artificial Intelligence: A Modern Approach," Russel, Norvig
- Daniel Weld, An introduction to least-commitment planning. Artificial Intelligence Magazine, 27--61, Winter 1994. [PDF]
- UCPOP Planner Web Page
Aug. 27: GraphPlan, Action Representations, Conditionals: [PDF]
- Ch. 11.4 "Artificial Intelligence: A Modern Approach," Russel, Norvig
- A. Blum and M. Furst, "Fast Planning Through Planning Graph Analysis", Artificial Intelligence, 90:281--300 (1997). [PDF]
- GraphPlan Web Page
- Corin R. Anderson et. al. Conditional Effects in GraphPlan In Proceedings of the 4th International Conference of AI Planning Systems, 1998. [PDF]
Sept. 1: Efficiency in Planning, Basics of Search: [PDF]
- Ch. 8 "Artificial Intelligence: A New Synthesis," Nilsson
- Ch. 3 "Artificial Intelligence: A Modern Approach," Russel, Norvig
Sept. 3: Heuristic Planning, Planning as Satisfiability: [PDF]
- B. Bonet and H. Geffner. Planning as Heuristic Search, Artificial Intelligence, Special issue on Heuristic Search. Vol 129 (1-2) 2001. [PDF]
- J. Hoffmann, FF: The Fast-Forward Planning System, AI Magazine, Volume 22, Number 3, 2001 [PDF]
- H. Kautz and B. Selman. Unifying SAT-based and Graph-based Planning. IJCAI 1999. [PDF]
- S. Richter and B. Westphal. The LAMA Planner: Using Landmark Counting in Heuristic Search. IPC 2009. [PDF]
Sept. 8: Classical Extensions and Applications: [PDF]
- NASA AIG (JPL) Webpage
- M. R-Moreno, J. Kurien What is Next in Autonomous Control Techniques? iSAIRIS 2008. [PDF]
- Further tutorials: 2000-2007 AAAI Planning Tutorial and 1999 IJCAI Planning Tutorial
- More Planner Links: Lots of Planners

Motion Planning

Sept. 10: Motion Planning: Complete Algorithms: [PDF]
- Ch. 2, 5.1 "Principles of Robot Motion," Choset et. al.
- Ch. 12.3.3-12.3.4, 6.2.4 "Planning Algorithms," S. LaValle
Sept. 15: Roadmap Planning: Visibility vs. Voronoi: [PDF]
- Ch. 5.1-5.4 "Principles of Robot Motion," Choset et. al.
- Ch. 6.2.3 - 6.2.4 "Planning Algorithms," LaValle
- Some Links to Voronoi Construction:
  F. Aurenhammer Voronoi diagrams: A survey of a fundamental geometric data structure ACM Computing Surveys, V.23 N.3, 1991. [PDF]
  D. Kirkpatrick Efficient Computation of Continuous Skeletons Symposium on Foundations of Computer Science, 1979. [PDF]
Sept. 17: Navigation Planning: Cells, Grids and Fields: [PDF]
- Ch. 6 "Principles of Robot Motion," Choset et. al.
- Ch. 6.1 - 6.3 "Planning Algorithms," LaValle
- S. Koenig, M. Likhachev: D*Lite. AAAI/IAAI 2002: 476-483 [PDF]
Sept. 22: Articulated Kinematics, Configurations: [PDF]
- Ch. 3.2-3.4 "Planning Algorithms," LaValle
- Wiegley, Lee and Goldberg's Configuration Space Applet
Sept. 24: Inverse Kinematics, Jacobians, Potential Fields: [PDF]
- Ch. 3 "Planning Algorithms," LaValle
- More detail: "Introduction to Robotics: Mechanics and Control," J. Craig
Sept. 29: Sampling-based Planning, (PRM) Probabilistic Roadmaps: [PDF]
- Ch. 5.6 "Planning Algorithms," LaValle
- Ch. 7.1 "Principles of Robot Motion," Choset et. al.
- Lin & Manocha's UNC Collision Detectors
- P. Jimenez, F. Thomas, C. Torras Collision Detection Algorithms for Motion Planning Robot Motion Planning and Control, J.P.Laumonnd Ed. [PDF]
- S Trenkel, R Weller, G Zachmann, A Benchmarking suite for static collision detection algorithms Proc of the Int. Conf. Computer Graphics 2007. [PDF]
Oct. 1: (RRT) Rapidly Exploring Random Trees, Path Improvement, Constraints: [PDF]
- Ch. 5 esp. 5.5 "Planning Algorithms," LaValle
- Ch. 7.2-7.3 "Principles of Robot Motion," Choset et. al.
- S. LaValle, J. Kuffner Randomized Kinodynamic Planning International Journal of Robotics Research, 20(5):378-400, 2001. [PDF]
- Stilman, M. Task Constrained Motion Planning in Robot Joint Space IROS 2007 [PDF]

Uncertainty and Dynamics

Oct. 8: Summary, Markov Systems, Matrix Inversion: [PDF]
- How to learn a model: "Machine Learning" T. Mitchell
- Useful reference for next week: "Reinforcement Learning" Sutton, Barto
Oct. 20: Final Project Proposal Presentations
- Another link to the Course Wiki
Oct. 22: Markov Decision Processes, Value Iteration: [PDF]
- Ch. 1-3 "Reinforcement Learning" Sutton, Barto (Introduction)
- Ch. 4.1-4.5 "Reinforcement Learning" Sutton, Barto (Value Iteration)
- D. Poole's Value Iteration Applet [Website]
Oct. 27: MDP: Applications & Algorithms: [PDF]
- Ch. 4.6 "Reinforcement Learning" Sutton, Barto (Policy Iteration)
- Ch. 11 "Reinforcement Learning" Sutton, Barto (Case Studies)
Oct. 29: Probability Primer: [PDF]
Nov. 3: Partially Observable MDP (POMDP): [PDF]
- L. Kaelbling, M. Littman, A. Cassandra, Planning and Acting in Partially Observable Stochastic Domains Artificial Intelligence, Volume 101, pp. 99-134, 1998. [PDF]
- A. Cassandra, A Survey of POMDP Applications. Presented at the AAAI Fall Symposium, 1998. [PDF]
- Tony Cassandra's POMDPs for Dummies
Nov. 5: POMDP vs. Kalman Filter: [PDF]
- P. Maybeck. "Stochastic Models Estimation and Control V.1" Ch. 1: Introduction [PDF]
- A. D'Souza Probabilistic Derivation of the Basic Kalman Filter, [PDF]
- Ch. 8.1-8.2 "Principles of Robot Motion," Choset et. al.
Nov. 10: Linear Control Primer: [PDF]
- Ch. 13.2 "Planning Algorithms," LaValle (Planning Book describes Linear Control)
Nov. 12: Linear Control: [PDF]
- Ch. 13.2 "Planning Algorithms," LaValle (Planning Book describes Linear Control)
- More detail: "Introduction to Robotics: Mechanics and Control," J. Craig
- Also: "A Mathematical Introduction to Robotic Manipulation," R. Murray, Z. Li & S. Sastry

Bridging Planning & Control

Nov. 17: Control Applied to Robot Trajectories: [PDF]
- "Introduction to Robotics: Mechanics and Control," J. Craig
- Khatib, O. "Inertial Properties in Robotic Manipulation: An Object-Level Framework" [PDF]
- Khatib, O. "Motion/Force Redundancy of Manipulators," Symposium on Flexible Automation [PDF]
  (Observe the distinction between gravity & full dynamic compensation)
Nov. 19: LQR Case Study:
- Ch. 8 "Mathematical Control Theory," E. Sontag
- Ch. 3 "Artificial Intelligence: A Modern Approach," Russel, Norvig
Nov. 24 to Final: Topics of Current interest in Planning

Assignments and Projects:

Project 1: Classical Sokoban Planner Due: 11:59pm Sept. 21
- Resources: hanoi.pddl, hanoi-domain.pddl
- Sokoban:
  Java Version
  Flash Version
  Odd Flash Version
- Suggested Planners (You are encouraged to try others):
  BlackBox Webpage, SatPlan Webpage
  FF Webpage, VHPOP Webpage
Project 2: Motion Planning Due: 11:59pm Oct. 28
- Resources: Download RST
Final Project Presentations: Dec. 11 (Fri) 12PM - NOON - Location TBD
- IEEE LaTeX Styles and Sample Documents and Page Requirements

Requirements:

This course has graduate (8803) and undergraduate (4803) sections. Both sections will participate in three group programming projects related to the three covered aspects of planning. The projects will be graded on algorithm implmementation, analysis and results for a total of 50% of the course grade.

Classical Planning (25%)
Motion Planning (25%)

In order to expose students to research in planning, the course will also have a final project that makes up 50% of the grade. This project will involve the design, implementation and validation of a planning algorithm resulting in a conference-style paper and presentation.

Robot arm planning and control.
Planning with abstractions.
Planning for a novel balancing platform.
Grasp planning for a robot hand.
Planning or control projects relevant to ongoing research.

8803 Graduate Projects:
Graduate students will work in groups on a project that is relevant to their research goals. The instructor will provide resources such as robot arm/hand hardware and existing algorithms to support this work. Furthermore, students are welcome to use resources or expand on active projects in their own research labs. Final decisions on topics will be made through discussion with the instructor.

4803 Undegraduate Reviews:
Undergraduates will take the role of reviewers for the projects. They will be exposed to research in planning and the peer-review process. Undergrads will be required to review project proposals, final projects, suggest alternative algorithms and find references that back up their claims. They will be graded based on the thoroughness of their reviews, understanding of the project topics and relevance of located references. Undergraduates are given the option to participate in the projects directly and be graded as graduate students.

CS 4803/8803: RIP - Robot Intelligence - Planning in Action TR 1:35-2:55, Bunger-Henry 380 Fall 2009

CS 4803/8803: RIP - Robot Intelligence - Planning in Action
TR 1:35-2:55, Bunger-Henry 380
Fall 2009