Robotics Basics - Hybrid Paradigm

Hybrid architectures are a combination of deliberative 20210711201454 and reactive 20210711201545 architectures. There is a middle layer between both parts that links them together and decides whether to choose one or another or a mixture of both to select the actions to perform.

PLAN; then SENSE -> ACT

Sensing is truly hibrid, as models are needed on planing and faster sensing is required for more reactive actions. Behaviors 20211217202757, rather than refer to purely reactive ones, now refer to also more complex ones, although this may cause some confusion (some architectures can use different terminologies to refer to different types of behaviors.)

The challenge for these architecture is to achieve the right compromise between these two layers. This is, indeed, a hard task, but this middle layer allows us to perform thing such as:

  • Deal with changes in the world when reactive architectures can’t proceed
  • Create a plan and follow it until reactive can’t proceed, and the replan.
  • Avoid constantly replaning by storing plans as intermediate actions
  • Pre-plan things in advance

To have a successful hybrid architecture, a feature that is mandatory is the ability to perform asynchronous processing and multithreading or multitasking. Heavy computational tasks are performed one one thread and quick process and update are done on another thread separately, to deal with environment changes, for example while planing the goal next to the actual one.

Hybrid architectures: differences and common elements

The main differences on hybrid architectures is how they deal with the following questions:

  • How to distinguish reaction and deliberation
  • How to organize responsibilities on deliberation
  • How to decide the overall behavior. We have the same as in reactive architecture, but the addition of a deliberation layer provides with more complex ways like fuzzy logic, voting or filtering.

Some common elements of this architecture are:

  • Sequencer: generates sequences of behaviors
  • Resource manager: allocates resources to behaviors
  • Cartographer: responsable of map information
  • Mission planner: interacts with humans and decides the next mission plan
  • Performance monitoring and problem solving agent

Styles of hybrid architectures

  • Managerial: Higher layers do high level planning that is passed to inferior layers that refine and gather resources to bottom layers. Higher layers can see results from lower layers, and lower layers can ask for more information/help to the top layers.
    • AuRA Architecture
    • Sensor Fusing Effects (SFX)
  • State-hierarchy: Organize architectures by scope of time knowledge: PAST, PRESENT and FUTURE, with 3 layers of knowledge that interact with each other.
    • 3T architecture
  • Model oriented: Use a global world model for sensing and actuation, using this model not only for planning but to pass sensing information to behaviors, like a virtual sensor.It is similar to the hierarchical paradigm, but the model is adapted to hybrid architectures
    • Saphira architecture
    • TCA architecture

Notes References

20210514183815 INDEX - Robotics

20211217202757 Robot behaviors

20210711200423 Robotics Basics - Control Paradigm

References

(Matariâc 2007) (Murphy 2000)

Matariâc, Maja J. 2007. The Robotics Primer. Intelligent Robotics and Autonomous Agents Series. Cambridge, Mass: The MIT Press.

Murphy, Robin. 2000. Introduction to AI Robotics. Intelligent Robotics and Autonomous Agents. Cambridge, Mass: MIT Press.