Mobile robots control architectures

Transcription

1 1 Mobile robots control architectures Dimitri Popov Universität Hamburg Fakultät für Mathematik, Informatik und Naturwissenschaften Department Informatik Integriertes Seminar Intelligent Robotics Juli 2010

2 Mobile robots control architectures Question of the day How to connect sensors and actuators? What to do next? 2

3 Outline Introduction 1 Introduction Summary & Conclusion 3

4 Outline Introduction 1 Introduction Summary & Conclusion 4

5 Definitions: Robotic Architecture Mataric, 1992 An architecture provides a set of principles for organizing control systems. However, in addition to providing structure, it imposes constraints on the way control problems can be solved. Arkin, 1998 Robotic architecture is the discipline devoted to the design of highly specific and individual robots from a collection of common software building blocks.. 5

6 General set-up Introduction Question of the day How to connect sensors and actuators? What to do next? We want could use Goals External Knowledge Internal states Algorithms We want to design The whole decision making process 6

7 Control architecture specification A mobile robot should be capable of Robust behavior Acting in a static or dynamic environment Learning capabilities A control system could be capable of Different goals Multiple sensors Modularity & Extensibility 7

8 Outline Introduction 1 Introduction 2 Subsumption architecture 3 4 Summary & Conclusion 8

9 Think hard, then act. Image source: Also known as Sense-(Model-)Plan-Act Popular in early robotics (from 1960) Sequential architecture 9

10 Purely deliberative architectures Knowledge, Data Base Mission Commands Localization Map Building Position Global Map Cognition Path Planing Environment Model Local Map Path Information Extraction and Interpretation Path Execution rol P ercep ti on Raw data Sensing Actuator Commands Acting Motio n Cont Real World Environment Image source: Introduction to Autonomous Mobile Robots, Siegwart '04 10

11 : Characteristics Deliberative Control...is a process of manipulating explicit representations of the world Essential: the planning process Required: Knowledge about the state of the world and plan Advantages High-level intelligence is possible Useful in a static environment Disadvantages Slow reacting to unexpected events Unusable in a highly dynamic environment 11

12 Don t think, react. Image source: Popular since 1980 Highly parallel architectures Basic idea: No memory, no states, just use reflexes 12

13 Generic reactive architecture Image source: The robotics primer, Mataric '07 13

14 Subsumption architecture Image source: Structure of one reactive element Element = final state machine + simple transforming function Inputs can be suppressed and outputs can be inhibited 14

15 Subsumption architecture Combinations of elements with layers Layering methodology for multiple elements is used. 15

16 : Characteristics Basic principles Tight sensing-action coupling ( as in animal reflexes) No explicit modeling Hierarchical, modular structure With asynchronous layers 16

17 : Characteristics Advantages Simplicity and robustness possible Timely responses in dynamic, unstructured worlds will be produced Never out of date Subsumption architecture: testing is relatively easy Disadvantages Robot will not be able to learn Sequencing is difficult 17

18 Think and act independently, in parallel. Image source: Trying to combine deliberative and reactive controls 18

19 : possible three-layer systems More Deliberative Image source: Behavior-based robotics, Arkin '98 ~ I I Level 2 Level 1 Level 0 Planner Deliberation ~ Projection Behavioral Advice Configurations Parameters [ ~~ ] : ~ =~ J More Reactive (A) (8) Figure 6.3 Typical deliberative /hierarchical planning strategies. (C) 19

20 : Characteristics...try to combine both deliberative and reactive architectures There are many quite different attempts to do so. Advantages Both intelligence, learning and fast reaction are possible at the same time Disadvantages It s hard to design a good middle layer. 20

21 Think the way you act. Image source: Can be seen as enchantment of reactive architectures Highly modular Reminds me of multi agent systems 21

22 Behavior-based control architectures an example Image Source: 22

23 Behaviors: definition and classification Behaviors......can be more complex than actions...achieve or maintain goals...are time-extended...can take inputs from actions and other behaviors and...send outputs to effectors and other behaviors...can be added at runtime...have to be on compatible time-scales 23

24 Behavior assembling Image source: The robotics primer, Mataric '07 24

25 Behavior-based control architectures Characteristics In behavior-based control Highly parallel system Alternative to hybrid systems Task-oriented decomposition Information is not centralized 25

26 Behavior-based control architectures Characteristics Advantages Fast real-time responses Modularity Possible to handle complex tasks Disadvantages Not feasible to use a global world model Design could result in too many modules 26

27 Outline Introduction Motor schema More architecture ideas 1 Introduction 2 3 Motor schema More architecture ideas 4 Summary & Conclusion 27

28 Motor schema More architecture ideas Motor schema Separate image sources: Behavior-based robotics, Arkin '98 28

29 Even more architectures Motor schema More architecture ideas Force control...as opposed to position control Agent-based architectures Defined protocols between agents Agent Coordination(Agent 1, Agent 2,..., Agent i ) 29

30 Outline Introduction 1 Introduction Summary & Conclusion 30

31 Summary & Conclusion A control structure is essential. There exist 4 basic ones: Deliberative Reactive Hybrid Behavior-based... as well as many specificities of them. The choice isn t always easy. 31

32 Thank you for your attention! 32

33 Appendix: subsumption architecture example Level 0: Avoid objects, run away if approached. Source: Brooks' first robot 33

34 Appendix: subsumption architecture example Level 0 and 1: Add wandering. Source: Brooks' first robot 34

35 Appendix: subsumption architecture example Level 0, 1 and 2: Add hallway following. Source: Brooks' first robot 35

36 Appendix: an agent architecture example State Communication Speech Vision Shared Memory Face/Expression Navigation Monitor Image source: Logical diagram of the REAPER Architecture. Each module takes inputs from and writes its outputs to the shared memory. The State module is the central controlling unit. 36

37 Appendix: Idea about the motor schema formalism ~NVIRONM~NT AL SENSORS E N V IR 0 NM E N T... Key: PS - Perceptual Schema PSS - Perceptual SUbschema MS - Motor Schema ES - EnvIronmental Sensor Image source: Behavior-based robotics, Arkin '98 37