AWS Services for Cloud Robotics Applications

Nowadays robots are equipped with a great number of sensors, in order to increase their level of autonomy. Since analyzing the data coming from such devices is very resource demanding, cloud robotics was born to offload some of the heavy tasks, such as computing, memory and storage, to the cloud. Furthermore the availability of computational resources and the effort to manage these infrastructures is minimal. However, not all the software can be unloaded to the cloud due to latency requirements and to maintain a local autonomy. In fact, the robot may be affected by disconnections during an operation and, therefore, a minimum level of autonomy must be guaranteed on-board to maintain an adequate level of safety, as well as avoiding it being stuck in place. In this thesis, an in-dept study of cloud technologies, such as Docker and Kubernetes, has been done. Subsequently some of the AWS' services have been analyzed and used in order to design a cloud architecture for robotic fleet management. This architecture is made by three different parts: the single robot development, the multi-robotic algorithm and a REST API to visualize the robots' status. Concerning the single robotic algorithm the attention is focused on AWS RoboMaker, a service that facilitates the creation of a robotic application through ROS and its simulation via the Gazebo GUI. The multi-robotic algorithm, instead, will be developed with AWS Cloud9, an IDE that supports different programming languages. Its containerized image will be uploaded on Amazon ECR, a container registry, and deployed in a Kubernetes' Pod, provided by Amazon EKS. The interaction between the Pod and the simulated robots will be possible via a NoSQL database offered by Amazon DynamoDB. Lastly, the different robot status are shown using a REST API, developed through the integration of a Python script, that retrieves the status from the database, AWS Lambda and Amazon API Gateway. In conclusion a different architecture, in case of availability of physical robots, will be exposed. Furthermore eProsima Fast DDS will be analyzed as a connection between ROS2 nodes and other nodes stored in a Kubernetes' Pod.