© sebastian thrun, CMU, 20001 16-899C Statistical Techniques In Robotics Sebastian Thrun and Geoffrey Gordon Carnegie Mellon University thrun.

© sebastian thrun, CMU, 2000 1

16-899C Statistical Techniques In Robotics

Sebastian Thrun and Geoffrey GordonCarnegie Mellon University

www.cs.cmu.edu/~thrunwww.cs.cmu.edu/~ggordon


notes

Pointer to Larry’s material


Administrative Information

Sebastian Thrun [email protected] Geoffrey Gordon [email protected] Web: http://www.cs.cmu.edu/~16899 Email list: [email protected] Time: Mon/Wed, 10:30-11:50am Location: NSH 3302 TA: n/a Appointments: send Email!



Goals

Enable you to program robots and embedded systems in a robust fashion

Enable you to understand the intrinsic assumptions in your robot software

Enable you to pursue original research in probabilistic robotics

Sway you into joining a young and fascinating research field: probabilistic robotics


What this course is not

Intro to robotics Little work Low on math


Course Schedule

LocalizationSept 4-16

MappingSept 30-Oct 16

Decision MakingOct 21-30

Multi-AgentNov 4-11

Advanced PerceptionNov 13-25


What You Should Do

Think

Think differently

Be critical

Come up with Original Research


What Is A Good Project

Mine Mapping Multi-Agent Control


Requirements

In teams of three:• Warm-up project (mobile robot localization)• Written assignment(s)• Research Project

Class Presence: all but two sessions (send me email) Quizzes (all but at most two) No exams


Your next tasks

Check out Web site• Read assigned paper• Download map+sensor data and program robot

localization algorithm

Send Sebastian mail with your name and names of team mates (for warm-up project)

Come to class on Sept 9th (10:30am-11:50am)






Five Sources of Uncertainty

EnvironmentDynamics

RandomAction Effects

SensorLimitations

InaccurateModels

ApproximateComputation


Trends in Robotics

Reactive Paradigm (mid-80’s)• no models• relies heavily on good sensing

Probabilistic Robotics (since mid-90’s)• seamless integration of models and sensing• inaccurate models, inaccurate sensors

Hybrids (since 90’s)• model-based at higher levels• reactive at lower levels

Classical Robotics (mid-70’s)• exact models• no sensing necessary



Rhino


Minerva


The CMU/Pitt Nursebot Initiative


People Detection

Mike Montemerlo


Learning Models of People

Maren Bennewitz


3D Mapping Result

With: Christian Martin


Multi-Robot Exploration


Mine Mapping (brand new)


What are interesting problems?

Mapping, automatic, manual, guided? Probabilistic localization, landmarks?, odometer!, Route planning, collision avoidance Mine Mapping?


How can we solve them?



Where Am I/?


Nature of Sensor Data: Uncertainty

Odometry Data Range Data



Warm-Up Assignment: Localization,Due Sept 23


Warm-Up Assignment: Localization


Warm-Up Assignment: Localization

© sebastian thrun, CMU, 20001 16-899C Statistical Techniques In Robotics Sebastian Thrun and Geoffrey Gordon Carnegie Mellon University thrun.

Documents

robotics sebastian thrun

sebastian mail

math slide

rhino slide

minerva slide

probabilistic robotics

original research slide

sensing necessary slide