Dynamic process migration in heterogeneous ROS-based environments

In distributed (mobile) robotics environments, the different computing substrates offer flexible resource allocation options to perform computations that implement an overall system goal. The AnyScale concept that we introduce and describe in this paper exploits this redundancy by dynamically allocating tasks to appropriate substrates (or scales) to optimize some level of system performance while migrating others depending on current resource and performance parameters. In this paper, we demonstrate this concept with a general ROS-based infrastructure that solves the task allocation problem by optimising the system performance while correctly reacting to unpredictable events at the same time. Assignment decisions are based on a characterisation of the static/dynamic parameters that represent the system and its interaction with the environment. We instantiate our infrastructure on a case study application, in which a mobile robot navigates along the floor of a building trying to reach a predefined goal. Experimental validation demonstrates more robust performance (around a third improvement in metrics) under the Anyscale implementation framework.