Consumer robotics: Fiction or reality?

Paolo Pirjanian, Chief Scientist

Evolution Robotics, Inc.
November 18, 2003

Overview

The challenges of consumer robotics

Key technology needs
Solutions provided by Evolution
Robotics
Building a viable B2B robot company

Challenges of consumer robotics

Challenge 1:
Challenge 2:

Challenge 3:
Challenge 4:
Challenge 5:
Ultimate:

Cost efficiency
Reliability (Real real-world
robotics)
Test and validation
Power efficiency
Miniaturization
Meeting customer
expectations.

Key technology needs

1.

Navigation
- Low-cost localization (and mapping)
- Reliable obstacle detection sensors/technologies
- Complete coverage
- Self-docking

2.

Human-Robot Interaction

- Voice recognition from a distance in noisy environments

- Human identification
- Gesture recognition
- Approach and follow (come here, follow me)
- Affective computing

Phase 1: Now we can do interesting stuff

(specialized hardware platform):
E.g., clean, patrol, deliver, gather information, entertain,

Key technology needs

3.

Flexible/generic hardware platform

- Flexible manipulation
- Pick and place
- Reach around and grab
- Stair climbing
- Variable terrain, Indoor/outdoor mobility

Phase 2: Now we can start doing some real interesting stuff

(flexible hardware platform):
E.g., Fetch, deliver, fix, assemble, cut, wash,
4.

Robot-Robot Interaction
- Communication
- Task allocation
- Coordination and Collaboration

5.

Learning

Phase 3: Sky is the limit.

Evolution Robotics

Technology Solutions and Services

Software Solutions
Applications & Services

Robot Kits

Embedded
Systems

Professional
Services
Training & Support
Platform Customization
Product Development
Application Development

Prototype Platforms

Technology solutions provided

by Evolution Robotics

Object recognition: Reliable object

recognition for navigation, HRI, servoing,
manipulation
vSLAM: Low-cost, vision-based localization
and mapping
Architecture: Cross platform middleware for
system integration

Object Recognition
Approach:

Extracts 1000 SIFT features of each

object. A very small subset of those
features with the right configuration is
required for identification of the object.

Estimation: Identification can provide

the name of the object and the full pose
of the camera with respect to the object.

Example applications

Visual servoing, navigation, docking

Edutainment: Reading book, visual

programming

Manipulation

SLAM

Sony Aibo uses ER Vision

Visual Simultaneous Localization &

Mapping

Fuses data from single USB camera and odom

Use sparse range of unique features. Dense

range is not necessary.

Takes pictures of unique locations to build a map

and uses those to estimate robots position

Accuracy of about +/- 25cm in x,y, and about 5

degrees in heading

vSLAM

Actual Path

Odometry

SLAM

Representative Images

ER Software Platform
(Integration middleware)

Architecture that runs on

many platforms
Has been embedded
Cross OS (3 OSs)
Highly independent of robot

(Near) Future work

Develop < $100 navigation system

SLAM
Path planning/execution
Obstacle detection/avoidance
Hazard detection/avoidance
Self-docking and charging

Contact information

www.evolution.com
Whitepapers, videos, press releases
Job descriptions
Sales

paolo@evolution.com