COMP 50: Autonomous Class tomorrow and 2/26 Intelligent Robotics - - PDF document

COMP 50: Autonomous Intelligent Robotics

Instructor: Jivko Sinapov

http://www.cs.tufts.edu/comp/50AIR/

Today: Planning Announcements Class tomorrow and 2/26

• Research Lecture homework: by the end of the

semester, attend and write a response to one of the following talks

Allison Chaney Tuesday 3/13 3:00 pm TOPIC: Human-Centered ML Aylin Caliskan Monday 2/26 3:00 pm TOPIC: Bias in AI and ML Derry Wijaya Thursday 2/22 3:00 pm TOPIC: Natural Language Processing Naomi Fitter Monday 3/12 9:30 am TOPIC: Human-Robot Cooperation

Class Projects

• Design Projects
• Infrastructure Projects
• Demo and Application Projects
• Combination of the above

Design Projects Design Projects (2)

• Design a mobile manipulator using mobile

bases, robot arms, and sensors that are currently on the market

Design in Simulation Demo Projects

• Message and object delivery
• “cocktail” robot (e.g., for a conference setting)
• Robot that looks for trash on the floor
• Robot that plays a game...
• ….

Infrastructure Projects

• Semantic mapping – add a layer of semantics to

the metric map that the robot currently has

• Symbolic planning – add symbolic actions and

STRIPS-like planning

• Enable 3D mapping – there are many packages

and libraries that exist but are largely experimental

• Enhance visual perception – train classifiers to

recognize specific objects of interest

“What can the robot do?”

Next Steps

• Team formation: By Friday, Feb. 22

– Look for a link to shared doc on Trunk

• Preliminary “Presentations” – Wed. Feb 28

– Lightning talks: < 5 min – 2-3 slides (enter link in shared doc) – Pitch 2-3 ideas for a class project to get feedback

from me and the rest of the class

Planning

[http://www.cs.colostate.edu/~cs540/spr2018/res_images/cs540logo.gif]

Teleoperation Teleoperation Teleoperation Teleoperation

Teleoperation Teleoperation Robotics Timeline Teleoperation vs Telepresence

• An early attempt to improve teleoperation was

to add more cameras / displays

• Telepresence aims for placing the operator in a

virtual reality that mimics the robot's surroundings

Telepresence Robots

http://www.pilotpresence.com/wp-content/uploads/2011/01/remote-presence-systemsv2.jpg

The need for (semi-) autonomy

How should autonomy be achieved and organized?

Robot Primitives

The Early Answer (1967): Sense-Plan-Act The Early Answer (1967): Sense-Plan-Act

Early Example of S-P-A Early Work on Planning

Early Work on Planning Early Work on Planning A More Realistic Example A More Realistic Example A More Realistic Example A More Realistic Example

Is INROOM(IT,R1) true or false? CONNECTS(D1,R1,R2)? INROOM(IT,R2)? IT

Representing Initial State

IT

Representing Goal State

IT

The “difference” table Logical Difference

• r

Finding the Plan Discussion

• How did you solve the problem?
• What are some limitations of planning with

STRIPS?

• Where do the predicates, operators, etc. come

from?

Towers of Hanoi with PDDL

[ https://s3.amazonaws.com/ka-cs-algorithms/hanoi-5-init.png ]

PDDL

• Editor: http://editor.planning.domains/
• Tutorial:

https://www.cs.toronto.edu/~sheila/2542/s14/A1/ introtopddl2.pdf

• Example PDDL files:

http://www.ida.liu.se/%7ETDDC17/info/labs/plann ing/strips/

Formulating the problem

• Step 1: specify predicates
• Step 2: specify actions with preconditions and

effects

3-Disk Hanoi Predicates

• clear(x):

true if x is clear, smaller(x1,x2): true if x1 is smaller than x2, ontop(x1,x2), disk(x1), peg(x1)

• Move Action variables: dx – the disk we are moving

– dy: destination disk/peg – du: the disk/peg under dx – Preconditions: smaller(dx, dy), clear(dy), clear(dx),

• ntop(dx,du)

– Effects: not clear(dy), not ontop(dx,du), clear(du),

• ntop(dx,dy),

Actions

• Action name and parameters:
• Preconditions:
• Effects:

Final Plam

PDDL Excercise

• Formulate the domain and problem for the

robot navigation example

Project Team-up Activity THE END