Page 1
© 2020, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Keynote: The role of cloud and open source software in the future of robotics
Roger Barga
C l o u d R o b o t i c s S u m m i t
General Manager
AWS Robotics
Brian Gerkey
CEO, Founder
Open Robotics
Page 2
Trends to track Role of cloud infuture of robotics
Page 3
What defines a robot?
A robot is an autonomous machine that is capable of sensing its
environment, that performs computations to make decisions, and
that performs actions in the real world.
ComputeSense Act
Page 4
Three distinct types of robots
Drones Robotic arms Ground mobility
Page 5
Mobile robotics
We are at an inflection point.
Expected growth in the use of mobile robots will increase by almost tenfold over the next two to three years.
333%
798%
Page 6
Mobile robotics
Source: IDTechEx
By 2030
70% of all mobile material handling equipment willbe autonomous
By 2023
It’s estimated that mobile autonomous robots will emerge as the standard for logistic and fulfillment processes
Page 7
The pull of economics
Sources: Economist Intelligence Unit; IMB; Institut fur Arbeitsmarkt und Berufsforschung; Int’l Robot Federation; US Social Security Data; McKinsey analysis
220
200
180
160
140
120
100
80
60
40
Labor costs
Robot prices
1990 1995 2000 2005 2010
Page 9
Top reasons for deploying Q. Please rank the top 5 reasons for deploying or planning to deploy
commercial service robots in your organization. N=550
Source: Commercial Service Robotics Survey IDC, July 2018
50.4
44.4
40.4
39.3
36.9
33.5
33.1
32.2
31.1
28.2
0 10 20 30 40 50 60
Page 10
3PLs
eCommerce/Retailers
Who’s using mobile robots?
Page 11
Other opportunities for mobile robots
Today just ~2% of mobile robots are automated
Robot platform
Fork trucks Tuggers
Pallet movers
Cross-dock
Pallet conveying
Unit load moves
Shipping/receiving
Replenishment
Process
Page 12
The future of mobile robots in logistics
Past Present Future?
Page 13
Takeaways
Economics is a significant driver.1
Improve worker productivity, offer customersnew services, and increase operational capacity.2
There’s consumer demand for new experiences.3
Page 14
Trends to track Role of cloud infuture of robotics
Page 15
AWS RoboMaker
Simulation Cloud extensions
for robotsFleet
management
Page 16
72 sensors
Low-end CPU
Cloud extensions
Customer story
Page 17
Customer story
Need
• Voice interface
• Real-time monitoring
• Live video streaming
Challenges
• Little expertise
• Limited local compute power
• Limited engineering resources
Solution
• AWS RoboMaker cloud extensions
Amazon Lex
Amazon Kinesis Video
Streams
Amazon Rekognition
Amazon CloudWatch
Amazon Polly
Page 18
Implementation
RobotCloud Extensions
Machine learning
Fleet management Diagnostics and logging
Over-the-air updates
Sockets
server
Real-time data
AWS
RoboMaker
ROS
AWS
Lambda
Amazon
S3
Amazon
CloudWatch
Amazon
Lex
Amazon
Polly
Amazon
Rekognition
AWS IoT
Greengrass
Page 19
Customer success
Results
• Built voice interface within hours
• Built live monitoring and alerting within days
• Built live video streaming within days
“It was a revelation seeing how easily cloud
connectivity could be accomplished with
[AWS] RoboMaker. We immediately realized
that we could use [AWS] RoboMaker to take
the next release of Lea to a higher level.”
Gabriel LopesControl and Robotics Scientist, Robot Care System
Page 20
AWS RoboMakersimulation
Zero infrastructure toprovision, configure,or manage
Run multiplesimulations in parallel
Auto-scale based onsimulation complexity
Pay-as-you-go
Page 21
Customer story
Need
• Test coverage for different floor layouts
• Test coverage for different scenarios, such as robot kidnap
• Improve code release speed
• Challenges
Challenges
• Costly and time consuming to test
• Limited test cases and coverage
• Late bug discovery
Solution
• Large-scale and automated testing using AWS RoboMaker simulation
Page 23
Customer success
Results
• 40 automated tests on each code commit
• 500 automated tests for each release candidate
• Much faster testing and release cycle(e.g., 1 hour versus 3 weeks for testing 70 robot kidnap scenarios)
Page 25
Reinforcement Learning for AWS DeepRacer
Page 26
Successful simulation to real transfer
Page 27
Role of the cloud in the future of robotics
Intelligent cloud services can enhance local processing on the robot and can improve performance over time.1
Simulation can be used to test application correctness, and ensure performance across a range of conditions. 2
Simulation, combined with imitation and reinforcement learning, can be used to program robot actuation. 3
Page 28
Robot software is hard
software
environment
sensorsactuators
Page 29
Develop and test in simulation
Page 30
Deploy the same code to hardware
Page 31
ROS
Robotics SDK
tools ecosystemcapabilitiesplumbing
Page 32
Technical Steering Committee
ROS 2
Goals
• Quality of design and implementation
• Validation, verification,and certification
• System reliability
• Flexibility in communication
• Real-time control and deterministic execution
• Support for smallembedded systems
Page 33
Latest release
Foxy Fitzroy—June 2020EOL—May 2023
Focus for 2020 Q3-Q4: Product readiness
Make ROS 2 more suitable for use
in production scenarios
Improve the out-of-box experience
for common use cases
Improve documentation
Address disparities between ROS 1 and ROS 2
ROS 2 status & roadmap
Page 34
Gazebo
Simulation as the best possible substitute for physical robots
GUISensors
Interfaces Cloud
Physics
Page 35
Ignition
Simulation libraries for reuse in other applications
ign-guiign-sensors
ign-rendering
ign-transport
ign-msgsign-fuel
ign-physics
ign-math
Page 36
Ignition/Gazebo status & roadmap
Latest release
Citadel—December 2020EOL—December 2024
Focus for 2020 Q3-Q4: Close the gap
Reduce feature disparity between
older and newer releases
Facilitate migration to new release
Improve documentation release
process and usability
Fully support Windows, Ubuntu, and macOS
Page 37
Use case: Humanoid supervised autonomy
Page 38
Use case: Multi-robot search and rescue
Page 39
Use case: Agile factory automation
Page 40
Simulation bottleneck:world authoring
Even with good tools and excellent artists, never enough hand-crafted worlds
Need automation to create massive sets of environments for testing
Page 41
World forge launch video