Textbook Outline

Mataric - Robotics Primer 

Chapter 13. Deliberative Control

  Deliberation – thoughtfulness in decision and action

  Shakey – used state of the art in machine vision as input to planner

  Planning  – taking into account all available knowledge of world & robot’s capabilities, producing a sequence of actions designed to achieve a specified goal.

  Searching (wrt planning)  – looking systematically through model of world to find goal state and derive a means of reaching it.

  Optimization – “an act, process, or methodology of making something (as a design, system, or decision) as fully perfect, functional, or effective as possible; specifically : the mathematical procedures (as finding the maximum of a function) involved in this” – Merriam-Webster

  Optimization criteria – the goals and constraints determining the possible optimization outcomes

  Cost of planning – when search space is large planning is expensive wrt to both time and space

  Real time – when a hardware or software system must respond within the time frame of its situated environment and the events of that environment, we say it must act in real time; often the time required for extensive planning exceeds the constraints of real time operation.

  SPA sequence – many deliberative, planner-based architectures involve these 3 steps performed in sequence – sensing, planning, acting (executing the plan).

  DRAWBACKS of deliberative (SPA) architectures:

–       Time-Scale

– combined input of sensors, internal models & representations, and set of possible actions creates very large planning search space

– time frame of situated environment does not match that required by slow planner

– Bottom line: Generating a plan for a real environment can be very slow

–       Space

– extensive & elaborate world models and computation with them requires a lot of memory

– Bottom line: Generating a plan for a real environment can be very memory-intensive

–       Information

– the more information about the world that is stored, the more information needs constant updating

– Bottom line: Keeping voluminous and elaborate world information updated is very time consuming

–       Use of Plans

– Any accurate plan is useful only if:

– environment does not change during execution in ways relevant to the plan

– robot knows state of world & of plan at all times

– robot’s effectors are accurate enough to execute plan faithfully

– Bottom line: Executing a plan is not trivial

  Some HISTORY:

–       Dissatisfaction

– during 60’s and 70’s drawbacks of deliberative approach became obvious

– during 80’s alternatives were proposed & developed

–       Outcome

– for most applications purely deliberative no longer used

– Exceptions are applications that:

– demand a great deal of advanced planning

– present a static environment

– have high degree of certainty in execution

– Example: robotic surgery

  CHAPTER SUMMARY:

– Deliberative called SPA = sense, plan, act

– Decompose control into separate and independent functional modules – sense-world, generate-plan, translate-plan-into-actions

– Execute modules sequentially – output of one becomes input of next

– Use centralized representation and reasoning

– Require extensive and slow computation for reasoning

– Encourage open-loop execution of generated plan