Track 6 Session 6_ 透過 AWS AI 服務模擬、部署機器人於產業之應用
•Download as PPTX, PDF•
4 likes•8,431 views
Recommended
More Related Content
What's hot
What's hot (20)
Similar to Track 6 Session 6_ 透過 AWS AI 服務模擬、部署機器人於產業之應用
Similar to Track 6 Session 6_ 透過 AWS AI 服務模擬、部署機器人於產業之應用 (20)
More from Amazon Web Services
More from Amazon Web Services (20)
Track 6 Session 6_ 透過 AWS AI 服務模擬、部署機器人於產業之應用
- 1. © 2020, Amazon Web Services, Inc. or its affiliates. All rights reserved. 透過 AWS AI 服務模擬、部署機器人 於產業之應用 T r a c k 6 | S e s s i o n 6 Bob Yeh Startup Solutions Architect Amazon Web Services
- 2. © 2020, Amazon Web Services, Inc. or its affiliates. All rights reserved. 2
- 3. What makes a robot? Perceive Plan Act … and repeat
- 4. Robotics use is accelerating in key industries Robotics is undergoing fundamental changes in collaboration, autonomous mobility, and increasing intelligence Source: IDTechEx By 2023, it’s estimated that mobile autonomous robots will emerge as the standard for logistic and fulfillment processes By 2030, 70% of all mobile material handling equipment will be autonomous Logistics Construction Consumer Home Energy and Utilities Retail Healthcare Oil and Gas Agriculture
- 5. © 2020, Amazon Web Services, Inc. or its affiliates. All rights reserved.
- 6. Design and develop robotics applications and functionality Agile development of robotics application requires software reuse and iterative development
- 7. AWS contributions to ROS 2 https://github.com/aws-robotics Quality of service (QoS) features for topics ROS 2 launch sandboxing extension Cross-compilation tools Secure-ROS 2 (SROS2) improvements Created and maintains rcpputils core package Runtime analysis tools address and thread sanitizers (Asan/Tsan)
- 8. © 2020, Amazon Web Services, Inc. or its affiliates. All rights reserved.
- 10. Challenges facing robotics developers Lack of virtual assets for simulation High cost of simulations at scale Lack of infrastructure to run simulations at scale Lack of infrastructure for deployment at scale Lack of security for deployment and update
- 11. AWS RoboMaker Easily simulate and deploy robotics applications at cloud scale Fleet management Test and verify Deploy and update Design and develop Cloud-based simulation Cloud-based simulation
- 12. Cloud-scale simulations Fully managed simulation infrastructure at cloud scale with pay-as-you-go pricing Multi-robot simulations for testing fleet operations Regression testing at cloud scale with CI/CD integration Machine learning model training
- 13. Use pre-built virtual 3D worlds provided out of the box, or bring your own Automatically provisioned, configured, and managed cloud infrastructure for Gazebo simulator Automatically scale based on simulation complexity Resource-based pay-as-you-go pricing at a minute granularity Cloud-scale simulations
- 14. Cloud-scale simulations Log files Log files 1 + Simulation world 1 Robot application Simulation job +Robot application Simulation world 2 Simulation job +Robot application Simulation world N Simulation job Log files 2 Log files N Metrics
- 15. Regression testing upon every code update and every software release Playing back recorded rosbags or running Gazebo- based simulations Large-scale and concurrent simulations triggered in a batch using AWS RoboMaker simulation APIs Integration with CI pipeline (Jenkins, Travis, AWS CodePipeline, etc.) Simulation use case 1 Regression testing at cloud scale with CI/CD integration
- 16. Problem • Limited test coverage for different floor layouts and scenarios • Costly and time-consuming to test • Late bug discovery in the field Solution • iRobot built a CI/CD pipeline for large-scale and automated testing using the AWS RoboMaker simulation service • More than 40 automated tests on each code commit and more than 500 automated tests for each release candidate Business benefits Much faster testing and release cycle (1 hour vs. 3 weeks for testing 70 complex localization scenarios) Case study: Regression testing
- 17. Use case 1: Regression testing with CI/CD integration AWS CodePipeline AWS CodeBuild Build and bundle code in ROS container Bundles stored in Amazon S3 AWS RoboMaker Create robot and simulation application AWS RoboMaker Launch simulations via AWS Lambda AWS RoboMaker Aggregate results from simulations Branch: Integration Git repository Clone on git hook (merge) AWS RoboMaker Simulation status check via AWS Lambda If test passed, merge with master Branch: Master (release) 1 2 3 4 5
- 18. Simulate multiple robots within the same environment Connect multiple simulations to a central fleet management software to test multi-robot orchestration Simulate inter-robot interactions or missions across robots Simulation use case 2 Multi-robot simulations for testing fleet operations
- 19. Problem • Bastian’s software solutions enable the orchestration of a fleet of robots • Software testing currently requires physical robots; practical limitation – can test only 8–10 robots in the lab Solution • AWS-enabled simulation of a multi-robot environment with 35+ robots, thus enabling testing without physical robots • AWS services used: AWS RoboMaker, AWS Lambda Business benefits Bastian Solutions is easily able to test applications for larger environments without having to stand up physical devices Case study 2: Multi-robot simulations
- 20. Rapidly generate trial data in simulation to train reinforcement learning model Train reinforcement learning model natively in the simulation or in Amazon SageMaker Run concurrent simulations to speed up training of a single model Simulation use case 3 Machine learning model training
- 21. Use case 3: Machine learning model training Use simulation to generate training data and test trained AI/ML models in simulation Reinforcement learning (RL) can be used to learn a control scheme in simulation Amazon SageMaker training Simulation environment Actions Observation Trained model RL agent Model updates Training data Observation Action Reward Deploy model Fleet management deployment
- 22. Deploy and update at cloud scale Ability to handle large fleet sizes Fully managed over-the-air update infrastructure at cloud scale Organization of robots by logical fleet Built-in security features
- 23. Register robots with AWS RoboMaker fleet management, and organize them into fleets Deploy a ROS application into a robot fleet securely with just a few clicks Conditional over-the-air (OTA) updates Fleet monitoring and alerting* Fleet deployment auto-rollback* * Coming soon Cloud-scale fleet management
- 24. Resources Tutorials and workshops Developer guide Try AWS RoboMaker today Regression testing at cloud scale with CI/CD integration Multi-robot simulations for testing fleet operations Machine learning model training in simulations Over-the-air deployment with cloud-scale fleet management aws.amazon.com/robomaker Blog
- 25. Thank you! © 2020, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Editor's Notes
- Title: Simulate and deploy robotics applications at cloud scale Target audience: Customers building robotics, including BDMs, DevOps, Developers, and Ops teams of robotics companies or partners and enterprises looking at robotics use cases Services covered: Primary: AWS RoboMaker, Secondary: AWS IoT, AWS IoT Greengrass, ML/Amazon SageMaker Description The business case for robotics and automation has never been more attractive. Enterprise adoption is increasing and consumer behavior is evolving, raising the demand for robotics of all types to assist with logistics, material movement, delivery, and equipment monitoring as well as the use of robots to simplify everyday tasks. Learn about the role that cloud services will play in shaping the future of robotics by allowing developers to partition functionality between their physical robot and the cloud, enhance the security of mission critical functionality, and enable the adoption of CI/CD pipeline practices to enhance the quality and speed of robotic app development. This block explains how customers are adopting AWS services to enable rapid testing through simulation, enhance robot functionality through cloud AI services, manage multi-robot fleets in production, and garner edge and performance data to provide valuable business insights. Resources for presenters: https://wisdom.corp.amazon.com/Pages/Solutions_Robotics.aspx BD: Vaibhav Phadnis, vphadnis@amazon.com Product Marketing: Dylan Locsin, dlocsin@amazon.com Product Management: Ray Zhu, rayzhu@amazon.com
- So what is a robot? <VET this statement Robots are like networked IoT devices – EXCEPT they also ACT and INTERACT with their environments, physically to do things for us> Robotics software has three broad components: Sense & Perceive the environment, Plan the actions, and Act on the plan. Within these, the planning and decision making is critical to add autonomy and intelligence to robotics applications.
- There are specific industries we see robotics playing an important role in. This includes logistics, construction, retail, hospitality, healthcare, agriculture, energy, oil and gas, facilities, and consumer home products. Each of these applications is demanding more autonomy and more intelligence from the robots operating in them. If you think about a robot in the home, we expect it to be social – to recognize us, talk to us, and interact with us. Or in a warehouse facility, we expect the robot to navigate and detect objects as its moving around the warehouse floor so it can effectively pick orders and move inventory.
- Let’s cover some basics: ROS – Robot Operating System. Most widely used software framework for teaching and learning about robotics – over 16 million .deb (Linux Debian) packages downloaded in 2018, a 400% increase since 2014. It’s actually not an OS, but is a middleware layer built to enable robotics applications Founded in Stanford labs over 10 year ago, now managed by the Open Source Robotics Foundation (OSRF) Global open-source community supports two products—Robot Operating System (ROS) and Gazebo Gazebo < DEFINITION adapted from their website > Robot simulation is essential for robotics development. Good simulation makes it possible to rapidly test algorithms, design robots, perform regression testing, and train AI system using realistic scenarios. Growing out of roots at the University of Southern California in 2002, Gazebo now offers the ability to accurately and efficiently simulate populations of robots in complex indoor and outdoor environments. It has a robust physics engine, high-quality graphics, and convenient programmatic and graphical interfaces. The OSRF continues development of Gazebo with support from a diverse, vibrant open source community.
- AWS is active in the open source communities, with our customers in the robotics space, especially ROS, and new ROS 2. Here are some things we’ve contributed to improve, harden and get ROS2 ready for production commercial usage. Cross-Compilation tools allows users to easily compile ROS2 for their target boards, from a build machine with a different architecture SROS2- Policy generation for securing nodes, generate keys for node communication and verification. Launch Sanboxing allows you launch nodes inside containers to monitor and restrict resource usage, think of micro services “cloud” techniques for ROS In ROS2 –D we solved: 4 memory leaks impacting production, 17 memory leaks impacting test and 2 data races impacting Fast RTPS Improved QA to detect memory and concurrency bugs ARMHF Support Rcpputils contains things like thread safety annocation macros, library discovery as well file system and type triats helpers
- <Need talking points>
- But developing robots with that level of autonomy and intelligence is difficult and time consuming. First, it requires machine learning expertise, as well as significant background in mechanical and software engineering. Most roboticists spend long periods of time prototyping as they evaluate all the different components necessary for autonomous functionality, including simulating cameras, sensors, and actuators on a physical dev kit while they optimize for both functionality and cost.. Due to the complexity of platforms, before a developer can start writing application code, they often have to spend hours or days to ensure the development environment is set up properly. They then have to download the robot OS software framework, different kinds of tooling, and ensure all those elements are working together properly. Additionally, not all developers working on robotic applications have access to robot hardware as this can be extremely costly and typically what is being used for prototyping is never the same as what goes to commercial production. Finally, there are many challenges with managing growing fleets and application updates for robots in the field.
- <Need talking points>
- <Need talking points> Highlight design and develop. ML model training is at design and develop stage. Simulation is not only for testing
- <Need talking points> <What is Gazebo vs. what is RoboMaker>
- <Need talking points>
- <Need talking points>
- <Need talking points>
- <Need talking points>
- <Need talking points>
- <Need talking points>
- <Need talking points>
- <Need talking points>
- <Need talking points>
- ROS and ROS 2.0 configured in the cloud Simulation service for parameter tuning Batch Simulation and CI/CD pipeline for regression testing Fleet Management provide over the air update capabilities to a robotic fleet. Cloud Extensions easily interface ROS with AWS services such as Amazon Kinesis Video Streams, Amazon Rekognition, and Amazon CloudWatch.