This document discusses the key aspects of planning and acting for intelligent agents. It begins with definitions of planning and outlines some of the assumptions of conventional AI planning, such as a static domain and directly executable actions. However, it notes that real-world planning and acting requires capabilities beyond these assumptions, like handling changes in the environment, interleaving planning and execution, and acting within a complex system. The document argues that future research in planning and acting should address these challenges to build systems that can effectively operate in the real world.
Sets, maps and hash tables (Java Collections)Fulvio Corno
Sets, maps and hash tables in the Java Collections framework
Teaching material for the course of "Tecniche di Programmazione" at Politecnico di Torino in year 2012/2013. More information: http://bit.ly/tecn-progr
From Declarative to Imperative Operation Specifications (ER 2007)Jordi Cabot
Declarative specifications are better but have ambiguity problems. Here I present my common sense -based approach to interpret declarative postconditions in a non-ambiguos way
Full paper: http://jordicabot.com/papers/ER07.pdf
(presented in the ER'07 conference)
Practical DevSecOps: Fundamentals of Successful ProgramsMatt Tesauro
From ONUG Fall 2022:
"Shift Left'' and automation have turned from ideals to meaningless buzzwords. Instead of riding the hype train, let's get real and cover practical and real-world examples taken from actual product security successes. Not every business is the same, neither will their DevSecOps program.
In this talk, I'll cover the fundamentals of common to successful DevSecOps programs as well as a grab bag of useful techniques to consider. These are lessons learned doing AppSec at a wide variety of companies including Rackspace, Pearson, a fortune 500 financial, Duo Security and Cognizant Healthcare. Bruce Lee said "Research your own experience. Absorb what is useful, reject what is useless, add what is essentially your own". The goal of this talk is to provide you with enough examples to build your own pragmatic and practical DevSecOps program or maybe absorb a new technique or two into your existing program.
Sets, maps and hash tables (Java Collections)Fulvio Corno
Sets, maps and hash tables in the Java Collections framework
Teaching material for the course of "Tecniche di Programmazione" at Politecnico di Torino in year 2012/2013. More information: http://bit.ly/tecn-progr
From Declarative to Imperative Operation Specifications (ER 2007)Jordi Cabot
Declarative specifications are better but have ambiguity problems. Here I present my common sense -based approach to interpret declarative postconditions in a non-ambiguos way
Full paper: http://jordicabot.com/papers/ER07.pdf
(presented in the ER'07 conference)
Practical DevSecOps: Fundamentals of Successful ProgramsMatt Tesauro
From ONUG Fall 2022:
"Shift Left'' and automation have turned from ideals to meaningless buzzwords. Instead of riding the hype train, let's get real and cover practical and real-world examples taken from actual product security successes. Not every business is the same, neither will their DevSecOps program.
In this talk, I'll cover the fundamentals of common to successful DevSecOps programs as well as a grab bag of useful techniques to consider. These are lessons learned doing AppSec at a wide variety of companies including Rackspace, Pearson, a fortune 500 financial, Duo Security and Cognizant Healthcare. Bruce Lee said "Research your own experience. Absorb what is useful, reject what is useless, add what is essentially your own". The goal of this talk is to provide you with enough examples to build your own pragmatic and practical DevSecOps program or maybe absorb a new technique or two into your existing program.
<p>From <a href="https://en.wikipedia.org/wiki/Site_reliability_engineering" target="_blank">Wikipedia</a>: Site reliability engineering (SRE) is a discipline that incorporates aspects of software engineering and applies that to operations whose goals are to create ultra-scalable and highly reliable software systems.<p>
<p>Over the past year Acquia has built their own SRE team to help their products and services scale with the demand of our growing number of customers. We wish to share our experience so that others are enabled to do the same and reap the rewards.</p>
<p>This presentation will discuss how the SRE team came about at Acquia, what achievements we have made so far, and the lessons we have learned along the way. We will then show the steps on how to introduce SRE to your workplace so you can deliver more reliable and scalable services to your customers! We will specifically cover:</p>
<ul>
<li>SRE's basic concepts and history from Google</li>
<li>The management support you will need to get started</li>
<li>Introducing the idea of service level objectives and error budgets</li>
<li>Operational Responsibility Assessments as a tool to measure risk</li>
<li>Creating a Launch Readiness Checklist to standardize and improve product launches</li>
<li>Finding ideal candidates for your SRE team</li></ul>
<p>The intended audience are software engineers, system administrators, and managers that have a desire to improve how they do their work and how their products/services perform.</p>
What Is Data-Driven Product Development by Aaptiv Senior PMProduct School
In this talk, we talked about how to implement a full cycle of collaborative, data-driven product development. Lisa explained how to use qualitative and quantitative research to make product decisions, and how to facilitate design and ideation workshops to encourage team problem solving. This talk went deep into real-world case studies from the digital fitness space.
Goodbye scope anxiety hello agility: Kanban implementation case study at amdocsYaki Koren
The main problem we were asked to help solve was scope instability: for years, the organization had been fighting to receive the entire scope upfront and, as time passed, this became less and less possible. Our customers’ (Telecommunication Service Providers) business required better responsiveness and flexibility. In the session we will describe our approach for the implementation: evolution. How we helped the managers to evolve and in parallel evolved our coaching practices. The session will show many examples of successful attempts at evolving and also of failures (which provide great opportunities for learning). The session should be a good kick-starter for lean agile implementation.
1. 1
Dana
Nau
and
Vikas
Shivashankar:
Lecture
slides
for
Automated
Planning
and
Ac0ng
Updated
1/23/15
This
work
is
licensed
under
a
CreaBve
Commons
AEribuBon-‐NonCommercial-‐ShareAlike
4.0
InternaBonal
License.
Chapter
1
Introduc0on
Dana S. Nau and Vikas Shivashankar
University of Maryland
2. 2
Dana
Nau
and
Vikas
Shivashankar:
Lecture
slides
for
Automated
Planning
and
Ac0ng
Updated
1/23/15
3. A systematic arrangement of
elements or important parts; a
configuration or outline: a seating
plan; the plan of a story.
4. A drawing or diagram made to
scale showing the structure or
arrangement of something.
5. A program or policy stipulating a
service or benefit: a pension plan.
plan n.
1. A scheme, program, or method
worked out beforehand for the
accomplishment of an objective:
a plan of attack.
2. A proposed or tentative project or
course of action: had no plans for
the evening.
Some
Dic0onary
Defini0ons
of
“Plan”
[a representation] of future behavior …
usually a set of actions, with temporal
and other constraints on them, for
execution by some agent or agents.
– Austin Tate, MIT Encyclopedia of
the Cognitive Sciences, 1999
3. 3
Dana
Nau
and
Vikas
Shivashankar:
Lecture
slides
for
Automated
Planning
and
Ac0ng
Updated
1/23/15
Introduc0on
Ø AI planning has made huge advances
over the past several decades
● Languages (PDDL, RDDL,
ANML) for specifying problems
● Mature technology for finding
plans in large state spaces
● Several well-known applications
Ø But less practical impact than one
might hope
Ø Less than several other areas of AI
● Machine learning
● Data mining
● Natural language processing
Ø Why?
4. 4
Dana
Nau
and
Vikas
Shivashankar:
Lecture
slides
for
Automated
Planning
and
Ac0ng
Updated
1/23/15
s s’Predict
Search
What
Does
Planning
Involve?
Ø Planning a course of action = Prediction + Search
● Predict what effects various
actions may have
● Search through alternative
sets of actions
● Look for plan that’s best
for the intended purpose
5. 5
Dana
Nau
and
Vikas
Shivashankar:
Lecture
slides
for
Automated
Planning
and
Ac0ng
Updated
1/23/15
“Conven0onal”
AI
Planning
Ø Focus: domain-independent techniques for
generating plans of action
Ø Given:
● Domain model (descriptions of
the states and actions)
● Planning problem
▸ Initial state s0
▸ Goal g
Ø Find a plan (sequence of actions)
▸ e.g., ⟨go-‐leM,
load,
go-‐right⟩
or a policy (set of state-action pairs)
▸ e.g., {(s0, go-‐leM), (s1, load), (s2, go-‐right)}
● Should be executable starting in s0
● Should produce a state that satisfies g
load
go-‐leM
go-‐right
loc1
loc2
loc1
loc2
loc1
loc2
loc1
loc2
loc2
s2
s1
s0
s3
6. 6
Dana
Nau
and
Vikas
Shivashankar:
Lecture
slides
for
Automated
Planning
and
Ac0ng
Updated
1/23/15
Key
Assump0ons
Ø Assumptions used in conventional AI planning
● Domain is static except for the planned
actions
▸ No other source of change in the world
● Planning is done offline
▸ Enough time to generate a complete plan
● Actions are directly executable primitives
▸ Planner doesn’t need to worry about how
they’ll be executed
● Actions can be represented by simple models
telling what they will do
▸ Required to be trivial to compute
• e.g., add/delete elements from sets
● Action models assumed correct and
complete
• e.g., include all possible outcomes
load
go-‐leM
go-‐right
loc1
loc2
loc1
loc2
loc1
loc2
loc1
loc2
s2
s1
s0
s3
7. 7
Dana
Nau
and
Vikas
Shivashankar:
Lecture
slides
for
Automated
Planning
and
Ac0ng
Updated
1/23/15
Interleaved
Planning
and
Execu0on
Ø Some good work on this topic
● But it still assumes actions are directly executable
Tracker
Target
Planner Executor
8. 8
Dana
Nau
and
Vikas
Shivashankar:
Lecture
slides
for
Automated
Planning
and
Ac0ng
Updated
1/23/15
Delibera0ve
Ac0ng
Requires
More
Ø Planner is part of a larger system: the actor
Challenges in P&A
‣ Open problems in addressed issues
‣ Several unaddressed issues
Monitoring
Goal reasoning
Acting Observing
Learning
Models, data &
knowledge bases
Environment
Mission
Criteria
Objectives
Monitoring
actions
Control
variables
Signals
Sensing
actionsFeedback
Feedback
Users
Planning
Robot’s platformExecution platform
9. 9
Dana
Nau
and
Vikas
Shivashankar:
Lecture
slides
for
Automated
Planning
and
Ac0ng
Updated
1/23/15
Example:
the
Key
to
Room
215
Dana Nau and Malik
Ghallab having breakfast
at a hotel near Paolo
Traverso’s home
Dana left before Malik was
finished, went to hallway,
started toward stairwell
Dana came back, told
Malik he had forgotten his
room key, picked it up
Went to hallway, saw
elevator was there, took it
to his floor
10. 10
Dana
Nau
and
Vikas
Shivashankar:
Lecture
slides
for
Automated
Planning
and
Ac0ng
Updated
1/23/15
Example:
the
Key
to
Room
215
Must
finish
tasks
in
my
room,
then
catch
a
bus
▸ Is
there
enough
Bme?
▸ Only
if
I
go
now
What
do
I
need
to
do?
▸ I’ll
need
my
room
key
▸ I
leM
it
on
the
table
▸ Go
back
and
get
it
Elevator
is
here
now
▸ My
room
is
2
floors
up
▸ If
I
use
elevator
instead
of
stairs,
I’ll
recover
some
lost
Bme
Dana Nau and Malik
Ghallab having breakfast
at a hotel near Paolo
Traverso’s home
Dana left before Malik was
finished, went to hallway,
started toward stairwell
Dana came back, told
Malik he had forgotten his
room key, picked it up
Went to hallway, saw
elevator was there, took it
to his floor
11. 11
Dana
Nau
and
Vikas
Shivashankar:
Lecture
slides
for
Automated
Planning
and
Ac0ng
Updated
1/23/15
Example:
the
Key
to
Room
215
Must
finish
tasks
in
my
room,
then
catch
a
bus
▸ Is
there
enough
Bme?
▸ Only
if
I
go
now
What
do
I
need
to
do?
▸ I’ll
need
my
room
key
▸ I
leM
it
on
the
table
▸ Go
back
and
get
it
Elevator
is
here
now
▸ My
room
is
2
floors
up
▸ If
I
use
elevator
instead
of
stairs,
I’ll
recover
some
lost
Bme
Dana Nau and Malik
Ghallab having breakfast
at a hotel near Paolo
Traverso’s home
Dana left before Malik was
finished, went to hallway,
started toward stairwell
Dana came back, told
Malik he had forgotten his
room key, picked it up
Went to hallway, saw
elevator was there, took it
to his floor
● Retrieve an abstract
temporal plan
Ø Do rough evaluation
from past experience
● Refine concurrently
with execution
Ø Predictive simula-
tion and monitoring
Ø Online replanning
and execution
● Event-driven comparison
of abstract plans
Ø Choose new plan
Ø Refine concurrently
with execution
12. 12
Dana
Nau
and
Vikas
Shivashankar:
Lecture
slides
for
Automated
Planning
and
Ac0ng
Updated
1/23/15
General
Characteris0cs
Ø Continual online planning
● Plans remained partial and abstract
until the details were needed
Ø Multiple levels of abstraction
● Go upstairs, go into room
▸ What route to take?
• How to open the door?
Ø Heterogeneous reasoning, heterogeneous models
● How much time will my tasks take?
● How to get to the stairwell?
● What to tell Malik when I go back to get my key?
Ø Is this just human reasoning, or is any of it needed in AI systems?
13. 13
Dana
Nau
and
Vikas
Shivashankar:
Lecture
slides
for
Automated
Planning
and
Ac0ng
Updated
1/23/15
Service
Robot
ungrasp
grasp
knob
turn
knob
maintain
move
back
pull
monitor
identify
type
of
door
pull
monitor
move
close
to
knob
open door
……
get out close door
respond to user requests
……
bring o7 to room2
go to
hallway
deliver
o7
…… … …
…
move to door
fetch
o7
navigate
to room2
navigate
to room1
Activities for a service robot to perform
Ø Multiple levels of abstraction
● Different state spaces and action spaces
● How to translate among them?
Ø Conventional AI planning can be
used for some of the tasks
● e.g., bring o7
to room2
Ø For others, need other
deliberation techniques
● e.g., how to open a door
Ø How to verify & maintain
consistency among different
action models?
14. 14
Dana
Nau
and
Vikas
Shivashankar:
Lecture
slides
for
Automated
Planning
and
Ac0ng
Updated
1/23/15
Harbor
Management
……
manage incoming shipment
unload unpack store
… ……
…… …
registration
manager
storage
assignment
manager
release
manager
booking
manager navigation
…
…
await order prepare deliver
storage area
C manager
storage area B
manager
storage area A manager
Ø Harbor management system based on
one for Bremen Harbor
● Imports/exports cars
Ø World changes dynamically
● Need to plan reactions
Ø Multiple levels of abstraction
● Top level can be planned offline
● The rest is online, based
on current conditions
Ø Operations are carried out by
different components that must
interact with each other
● Need operational models of the
components
● On-the-fly synthesis of automata
to control the interactions
15. 15
Dana
Nau
and
Vikas
Shivashankar:
Lecture
slides
for
Automated
Planning
and
Ac0ng
Updated
1/23/15
General
Characteris0cs
Ø Same characteristics as in the summary of the “Room 215” example
Ø Continual online planning
● Plans remain partial and abstract
until the details were needed
Ø Multiple levels of abstraction
● Manage shipment
▸ Storage assignment
• Booking manager
Ø Heterogeneous reasoning, heterogeneous models
● Different kinds of action models
● Interactions among processes
16. 16
Dana
Nau
and
Vikas
Shivashankar:
Lecture
slides
for
Automated
Planning
and
Ac0ng
Updated
1/23/15
Focus
of
the
Book
(and
the
Course)
Ø Planning
Ø Acting
Ø How to combine them
Deliberation
components
Execution platform
Commands Percepts
Other
actors
Objectives
Messages
External World
SignalsActuations
Actor
Deliberation components
Execution platform
Planning
Acting
Queries
Plans
17. 17
Dana
Nau
and
Vikas
Shivashankar:
Lecture
slides
for
Automated
Planning
and
Ac0ng
Updated
1/23/15
Outline
of
the
Book
Ø Chapter 1: Introduction
● That’s what we’re covering right now J
Ø Chapter 2: Deliberation with deterministic models
● The “conventional” AI planning techniques mentioned earlier
▸ Sometimes called “classical” AI planning
● Action models, planning algorithms
● Integrating them with acting
Ø Chapter 3: Deliberation with refinement methods
● Hierarchical decomposition of larger problems in to smaller ones
● Reactive execution algorithm
● Planning algorithm
● Combining them
Ø Chapter 4: Deliberation with temporal domain models
● Modeling the time required to perform an action
● Algorithms for planning and acting
18. 18
Dana
Nau
and
Vikas
Shivashankar:
Lecture
slides
for
Automated
Planning
and
Ac0ng
Updated
1/23/15
Outline
of
the
Book
Ø Chapter 5: Deliberation with nondeterministic domain models
● Action models in which the actions have multiple possible outcomes
● Different kinds of solutions (safe/unsafe, cyclic/acyclic)
● Planning techniques (determinization, model checking)
● Acting techniques (I/O automata, synthesis of controllers)
Ø Chapter 6: Deliberating with probabilistic domain models
● Actions with multiple possible outcomes, and probabilities of those outcomes
● Stochastic Shortest Path problems (generalization of MDP problems)
● Planning algorithms (policy and value iteration, other more advanced
approaches)
● Acting with probabilistic models
Ø Chapter 7: Other deliberation functions
● Perceiving, monitoring, goal reasoning, interaction, learning, hybrid models,
ontologies
Ø Chapter 8: Conclusion
19. 19
Dana
Nau
and
Vikas
Shivashankar:
Lecture
slides
for
Automated
Planning
and
Ac0ng
Updated
1/23/15
Any
Ques0ons?