Chapter 12 Planning and Acting in the Real World Ch 11 Issue: STRIPS - PowerPoint PPT Presentation

Chapter 12 – Planning and Acting in the Real World Ch 11 Issue: STRIPS talks about what actions to do, but not about how long it takes or even when actions should occur. STRIPS is a bit like playing dominoes… Considering time is more realistic (which STRIPS also doesn’t do). Common planning need � Job shop scheduling. Example – Auto assembly. Fig 12-1 in book. Assembling 2 cars.

[x,y] = earliest & latest possible start times. Critical path � Duration � Fig 12-2 A solution to the auto job-shop scheduling problem.

Resource: Fig 12-4 A solution to the job shop scheduling problem with resources shown. Resources are reusable or consumable . Resources often are aggregated , like Inspectors(2). Sample algorithm used (p. 421) – “Minimum slack” – Considers the unscheduled actions that have had all their predecessors scheduled and schedules the one with the least slack for the earliest possible start. A “greedy” algorithm. Book takes a “plan first, schedule later” approach (p. 422). In “real world” the planning part may be done by humans, the scheduling by an algorithm.

Better for complex plans and schedules – Hierarchical decomposition (p. 422) into a “Hierarchical Task Network” (HTN): At each “level,” only a small number of individual planning actions, then descend to lower levels to “solve these” for real. Fig 12-5 One possible decomposition for the BuildHouse action. At higher levels, the planner ignores “internal effects” of decompositions (p. 425). But these have to be resolved at some level…

Fig 12-7 Decomposition of a high-level action within an existing plan.

Nothing’s perfect, including HTN planning (p. 427) – 1. Need to rule out recursion – so that HTN plans are of finite length and can be enumerated. 2. Need to bound the length of solutions. Avoid loops. 3. Adopt the hybrid approach that combines POP and HTN planning. Partial-order-planning (Ch 11) decides whether a plan exists. Then plans can be sophisticated, like --

O Henry seems perplexed by our planner! Fig 12-8 The Gift of the Magi problem – inconsistent states decomposed into a consistent solution! Right – The O Henry short story made into a play…

This was all “classical planning” so far. Tomorrow – We’ll talk about planning when you have “issues” like incomplete and incorrect information. (pp. 431 +) Some of the variations – • Sensorless planning – using coercion • Conditional planning • Execution monitoring and replanning • Continuous planning

Consider the book’s “painting a table and chair the same color” plan, with these planning variations – • Sensorless planning – using coercion • Conditional planning • Execution monitoring and replanning • Continuous planning

Conditional The “double Murphy” vacuum scenario. Planning Or, “What’s the opposite of Eureka!”? Fig 12 – 9 The first two levels of the search tree for the “double Murphy” vacuum cleaner world. (Cleaner sometimes deposits dirt when it moves to a clean destination square, and sometimes deposits dirt when Suck is applied to a clean square.)

Fig 12 – 11 The first level of the search graph for the “triple Murphy” vacuum cleaner world. (Sometimes the vacuum cleaner also won’t move.)

Conditional Planning Multiplies the possibilities! In Partially Observable Environments Fig 12 – 12 Part of the AND-OR graph for the “alternate double Murphy” vacuum cleaner world. (You can leave dirt behind when you move.)

Execution Monitoring and Replanning

Fig 12 – 14 Before execution, the planner comes up with a plan … then repaired. Above – In a replanning exercise, little bots at Stanford maneuver out of the way of an incoming obstacle. From robotics.stanford.edu/~dyhsu/ projects/rand/kinodyn.html

Continuous Planning Here the planning agent persists indefinitely… Fig 12 – 15 The sequence of states as the continuous planning agent tries to reach the goal state. Left – Helsinki metro, where maintenance planning is a continuous process.

Figs 12 – 16 / 18 Successive plans constructed by the continuous planning agent.

Figs 12 – 19 / 21 Successive plans constructed by the continuous planning agent, cntd.

Multiple agents… As in games, only they can cooperate, too! But, How do you coordinate the agents’ plans?

Coming up – Tonight – Erik’s in the lab 7 – 9 PM. New version of his .classes put out on the web last night – bug fixes. Thurs & Fri – Ch 13 – Uncertainty Fri, first ½ hour – Quiz on Ch 10, 11, 12, 13 Fri night – Games due, to run over the weekend… But, we’ll vote on moving this due date to Monday night… To be verified with confirming e-mail. Uncertainty – When will the flower open?