Machine Learning of Robot Assembly Plans

The study of artificial intelligence (AI) is indeed a strange pursuit. Unlike most other disciplines, few AI researchers even agree on a mutually acceptable definition of their chosen field of study. Some see AI as a sub field of computer science, others see AI as a computationally oriented branch of psychology or linguistics, while still others see it as a bag of tricks to be applied to an entire spectrum of diverse domains. This lack of unified purpose among the AI community makes this a very exciting time for AI research: new and diverse projects are springing up literally every day. As one might imagine, however, this diversity also leads to genuine difficulties in assessing the significance and validity of AI research. These difficulties are an indication that AI has not yet matured as a science: it is still at the point where people are attempting to lay down (hopefully sound) foundations. Ritchie and Hanna [1] posit the following categorization as an aid in assessing the validity of an AI research endeavor: (1) The project could introduce, in outline, a novel (or partly novel) idea or set of ideas. (2) The project could elaborate the details of some approach. Starting with the kind of idea in (1), the research could criticize it or fill in further details (3) The project could be an AI experiment, where a theory as in (1) and (2) is applied to some domain. Such experiments are usually computer programs that implement a particular theory.

1 Introduction.- 1.1. Machine Learning.- 1.2. Robotics.- 1.2.1. Why Can’t Robbie Learn?.- 1.2.2. Teach-By-Guiding Systems.- 1.2.3. Robot Programming Systems.- 1.2.4. Myopia on the Road to Intelligent Robots.- 1.3. About the Book.- 1.3.1. Organization.- 1.3.2. On the Use of the $ Symbol.- 2 Scenario.- 2.1. Preliminaries.- 2.1.1. The Widget.- 2.1.2. Moving the Robot Arm.- 2.2. Specifying the Problem.- 2.2.1. Describing the Initial State.- 2.2.2. Specifying the Goal State.- 2.3. Attempting to Solve the Problem.- 2.4. Observing the Expert’s Plan.- 2.5. Generalizing the Solution.- 2.6. Solving the Same Problem After Learning.- 2.7. Solving Similar Problems After Learning.- 3 Explanation-Based Learning.- 3.1. Similarity-Based Learning.- 3.1.1. Applying SBL to Classification Tasks.- 3.1.2. Applying SBL to Problem-Solving Tasks.- 3.2. Learning-Apprentice Systems.- 3.3. Explanation-Based Learning.- 3.4. A Prototypical EBL System.- 3.4.1. The Performance Element.- 3.4.2. The Learning Element.- 3.4.2.1. The Understander.- 3.4.2.2. The Generalizer.- 3.5. Issues for EBL Systems.- 4 The Arms World.- 4.1. Characterizing the Robot World.- 4.1.1. The Pieces.- 4.1.2. The Workspace.- 4.1.3. The Robot Arm.- 4.1.4. The Robot World Domain Theory.- 4.2. Simulating the Robot World.- 4.2.1. The Solid Modeler.- 4.2.2. The Emulator.- 4.2.2.1. Moving the Robot Arm.- 4.2.2.2. Modeling Robot/Piece Interactions.- 4.2.3. The History Mechanism.- 5 Learning And Problem Solving.- 5.1. Knowledge Representation.- 5.1.1. The Schema System.- 5.1.1.1. State Schemata.- 5.1.1.1.1. Constraint Schemata.- 5.1.1.1.2. Joint Schemata.- 5.1.1.2. Operator Schemata.- 5.1.2. The Database Mechanism.- 5.1.2.1. State Schema Validation.- 5.1.2.2. Caching Valid State Schema Instances.- 5.1.2.3. Database Parallelism.- 5.2. The Performance Element.- 5.2.1. The Design Phase.- 5.2.2. The Planning Phase.- 5.3. The Learning Element.- 5.3.1. The Understander.- 5.3.1.1. Specifying the Initial State.- 5.3.1.2. Emulating the Input Sequence.- 5.3.1.3. Building the Causal Model.- 5.3.1.3.1. Predictive Understanding.- 5.3.1.3.2. Nonpredictive Understanding.- 5.3.1.3.3. The Schema-Activation Mechanism.- 5.3.2. The Generalizer.- 5.3.2.1. The Verification Process.- 5.3.2.1.1. Known Physical Joint Schema.- 5.3.2.1.2. New PhysicalJoint Schema.- 5.3.2.2. Extracting the Explanation.- 5.3.2.3. Building a New Operator Schema.- 5.3.2.4. Meeting the Retention Criteria.- 5.3.2.5. Integrating Newly Acquired Schemata.- 5.3.2.6. Meeting the Replacement Criteria.- 6 The Arms Implementation.- 6.1. A Note About the Implementation Language.- 6.2. Optimization Tools.- 6.2.1. $MatchMixin.- 6.2.2. $LazyCopy.- 6.3. Implementing the Solid Modeler.- 6.3.1. Homogeneous Coordinates.- 6.3.2. $WorkSpace.- 6.3.3. $PositionedObject.- 6.3.4. $Piece.- 6.3.5. $Primitive.- 6.3.6. $Block, $Cylinder.- 6.3.7. $Surface.- 6.3.8. $PlanarSurface, $CylindricalSurface.- 6.3.9. $Hole.- 6.3.10. $Arm.- 6.4. Implementing the Graphics Subsystem.- 6.4.1. $View.- 6.4.2. $Segment.- 6.5. Implementing the Schema System.- 6.5.1. $Schema.- 6.5.2. $StateSchema.- 6.5.2.1. $ConstraintSchema.- 6.5.2.2. $JointSchema.- 6.5.2.2.1. $DegreeOfFreedom.- 6.5.2.2.1.1. $PrismaticDOF, $RevoluteDOF.- 6.5.2.2.2. $CylindricalJoint.- 6.5.2.2.3. $RigidJointA.- 6.5.3. $OperatorSchema.- 6.5.3.1. $PrimitiveSchema.- 6.6. Implementing the Top Level.- 6.6.1. General Description of $Episode.- 6.6.2. Implementing the History Mechanism.- 6.6.3. Implementing the State Schema Database.- 6.6.4. Implementing the Planner.- 6.6.5. Implementing the Understander.- 6.6.6. Implementing the Verifier.- 6.6.7. Implementing the Generalizer.- 7 Scenario Revisited.- 7.1. Attempting to Solve the Problem.- 7.2. Observing the Expert’s Plan.- 7.3. Verifying the Solution.- 7.4. Generalizing the Solution.- 7.4.1. A More General New Schema.- 7.4.2. A More Operational New Schema.- 7.5. Solving the Same Problem After Learning.- 7.6. Solving Similar Problems After Learning.- 7.7. Observing Similar Problems After Learning.- 8 Summary And Future Work.- 8.1. Relation to Other Work.- 8.1.1. STRIPS.- 8.1.2. MA.- 8.1.3. LEAP.- 8.1.4. ODYSSEUS.- 8.1.5. PRODIGY.- 8.2. Extensibility of ARMS.- 8.2.1. The Solid Modeler Problem.- 8.2.2. Reasoning with Uncertainty.- 8.2.3. The Operator/State Problem.- 8.2.4. The Temporal Reasoning Problem.- 8.3. Future Research Directions.- 8.3.1. Frame Selection Problem.- 8.3.2. Other Explanation Construction Methods.- 8.3.3. When and What to Learn.- 8.3.4. When and What to Forget.- 8.3.5. Refining Existing Knowledge.- 8.3.6. Learning Control Knowledge.- 8.3.7. Extending Imperfect Domain Theories.- 8.3.8. Execution Monitoring and Plan Revision.- 8.3.9. Dealing with Multiple Plans.- 8.4. Conclusions.- Appendix A Solid Modeling Systems.- Appendix B Schema Semantics.- Appendix C A Simpler Example.- C.1. Describing the Initial State.- C.2. Attempting to Solve the Problem.- C.3. Observing the Expert’s Plan.- C.4. Verifying the Solution.- C.5. Generalizing the Solution.- C.6. Solving the Same Problem After Learning.- C.7. Solving Similar Problems After Learning.- C.8. Observing Similar Problems After Learning.- Appendix D A More Complex Example.- D.1. Describing the Initial State.- D.2. Attempting to Solve the Problem.- D.3. Observing the Expert’s Plan.- D.4. Verifying the Solution.- D.5. Generalizing the Solution.- D.5.1. A More General New Schema.- D.5.2. A More Operational New Schema.- D.6. Solving the Same Problem After Learning.- D.7. Solving Similar Problems After Learning.- Appendix E Performance Considerations.- E.1. Learning Episode 1.- E.2. Problem-Solving Episode 1.- E.3. Problem-Solving Episode 2.- E.4. Problem-Solving Episode 3.- E.5. Problem-Solving Episode 4.- E.6. Learning Episode 2.- E.7. Learning Episode 3.- E.8. Problem-Solving Episode 5.- E.9. Problem-Solving Episode 6.- Appendix F Built-in Schemata.- F.1. State Schemata.- F.1.1. Joint Schemata.- F.1.2. Degree of Freedom Schemata.- F.1.3. Constraint Schemata.- F.2. Operator Schemata.- F.2.1. Primitive Operator Schemata.- References.