Welcome to M/W 1:30 2:20 Lectures, discussions (EEB 045) CSE 571 - - PowerPoint PPT Presentation

1

SA-1

Welcome to CSE 571 Probabilistic Robotics

Instructor: Dieter Fox Teaching Assistant: Arun Byravan

Probabilistic Robotics 10/1/15 2

Organization

• M/W 1:30 – 2:20
• Lectures, discussions (EEB 045)
• Homework, project
• Readings:
• Papers
• Chapters from Probabilistic Robotics
• Web page:
• http://www.cs.washington.edu/571

Probabilistic Robotics 10/1/15 3

Goal of this course

• Provide an overview of problems /

techniques in robotics

• Deep understanding of estimation in

dynamic systems

• Probabilistic models
• Inference, learning
• Hands-on experience

Probabilistic Robotics 10/1/15 4

Actions Control system Sensor data World model

High-level View on Robot Systems

2

Probabilistic Robotics 10/1/15 5

Robotics Yesterday

Probabilistic Robotics 10/1/15 6

Current Trends in Robotics

Robots are moving away from factory floors to

• Entertainment, toys
• Homes, hotels (personal robotics)
• Medical, surgery
• Industrial automation

(mining, harvesting, warehouses, …)

• Hazardous environments

(space, underwater, battlefields, …)

• Roads

Probabilistic Robotics 10/1/15 7

Minerva (CMU + Univ. Bonn, 1998)

Probabilistic Robotics 10/1/15 8

Architecture of the Control System

3 RoboCup: Integrated System Research

• Focus on addressing all problems at once
• Hardware development
• Perception
• Low level control
• High level planning and decision making
• Multi robot systems

Probabilistic Robotics 10/1/15 9 Probabilistic Robotics 10/1/15 10

RoboCup-99, Stockholm, Sweden

RoboCup Small Humanoid League

Probabilistic Robotics 10/1/15 11

RoboCup: Midsize League

Probabilistic Robotics 10/1/15 12

4 RoboCup Rescue

Probabilistic Robotics 10/1/15 13

DARPA Urban Challenge 2007

Probabilistic Robotics 10/1/15 14

Google Self-Driving Car

10/1/15 Probabilistic Robotics 15

Control: BigDog

Probabilistic Robotics 10/1/15 16

5 Cheetah

Probabilistic Robotics 10/1/15 17 Probabilistic Robotics 10/1/15 18

Boston Dynamics Cheetah

Probabilistic Robotics 10/1/15 19

DARPA Robotics Challenge 2015

Probabilistic Robotics 10/1/15 20

6

Probabilistic Robotics 10/1/15 21

Humanoids: Honda P2

Honda P2 ‘97

Getting out of Car

Probabilistic Robotics 10/1/15 22

Drilling Hole

Probabilistic Robotics 10/1/15 23

Current Research Trends / Topics

• Manipulation of everyday objects
• Complex household tasks (cooking,

cleaning, …)

• Kinect for object detection, 3D

mapping, tracking, interaction

• Human robot interaction
• Machine learning for control, imitation

learning, recognition

• Deep learning

Probabilistic Robotics 10/1/15 24

7

25

Course Outline

Week Content

HW / Project

#1 Introduction

Probabilistic Models / State Estimation

#2 Bayesian state estimation / filtering #2 Motion and sensor models, Gaussian processes HW 1: GP modeling

Filtering (localization, mapping)

#3 / 4 Robot localization: grid, particle filters, EKF, UKF HW2: Filtering #5 / 6 Map building: EKF-SLAM, Fast-SLAM, RGBD

Planning / Control

#6 / 7 / 8 Path planning, exploration, MDPs, POMDPs Project #9 Reinforcement learning, inverse RL

Other Topics

#10 Object detection and tracking

10/1/15 Probabilistic Robotics

Sample-based Localization (sonar)

10/1/15 26 Probabilistic Robotics

Graphical Model Representation

• f Localization Problem

10/1/15 27 Probabilistic Robotics

Mapping with Laser Scanners

Probabilistic Robotics 10/1/15 28

8 Mapping with Kinect

Probabilistic Robotics 10/1/15 29

SLAM: Simultaneous Localization and Mapping

30 10/1/15 Probabilistic Robotics

m x z u x z u

2 2

x z u

...

t t

x

1 1 1 t-1 Probabilistic Robotics 10/1/15 31

Structured Estimation Localization and Ball Tracking

k-2

b

k-1

b

k

b r

k-2

r

k-1

r

k

z

• k-2

z

k-1

z

k

u

k-2

u

k-1

z

k-1

z

k

z

k-2 k

m

k-1

m

k-2

m

Ball observation Ball location and velocity Ball motion mode Map and robot location Robot control Landmark detection

• Ball tracking
• Robot localization

l l l b b b

10/1/15 32 Probabilistic Robotics

9

Articulated Tracking (42 DOF)

10/1/15 Probabilistic Robotics 33

Gaussian Process Sensor Model for WiFi Signal Strength

• Non-parametric

regression

• GP regression
• continuous locations
• smooth interpolation
• uncertainty estimates

34

Mean Variance

10/1/15 Probabilistic Robotics

Tracking Example

35 10/1/15 Probabilistic Robotics

RL with GP Dynamics Models: PILCO (Probabilistic Inference for Learning Control)

36 10/1/15 Probabilistic Robotics

10

Pedestrian Trajectory Prediction

[Ziebart-Bagnell-etal]

10/1/15 Probabilistic Robotics 37

Pedestrian Trajectory Prediction

¡ Inverse optimal control: Learn cost function that

explains human behavior; use that to estimate goal

38 10/1/15 Probabilistic Robotics

Planning for Manipulation

Probabilistic Robotics 10/1/15 39