QoS metrics-in-the-loop for endowing runtime self-adaptation to robotic software architectures

The design of robots capable of operating autonomously in changing and unstructured environments, requires using complex software architectures in which, typically, robot engineers manually hard-code adaptation mechanisms allowing the robot to deal with certain situations. As adaptation is closely related with context monitoring, deliberation and actuation, its implementation typically spreads across several architecture components. Therefore, fine-tuning or extending the adaptation logic (e.g., to cope with new contingencies not foreseen at design-time) results in a very expensive and cumbersome process. This paper proposes a novel approach to deal with self-adaptation based on modeling behavior variability at design-time so that the robot can configure it at runtime, according to the contextual information only then available. This approach is supported by a model-based framework allowing robotic engineers to specify (1) the robot behavior variation points (open decision space); (2) the internal and external contextual information available; and (3) the non-functional properties (e.g. safety, performance, or energy consumption) in terms of which the robot Quality-of-Service (QoS) will be measured. Then, from these models, the framework will automatically generate the runtime infrastructure allowing the robot to self-adapt its behavior to achieve the best QoS possible according to its current context. The framework has been validated in two scenarios using two different well-known robotic software architectures.

Automated summary

This paper proposes an approach to self-adaptation in robots by modeling behavior variability at design-time and allowing the robot to configure its behavior at runtime based on contextual information. This approach is supported by a model-based framework that allows robotic engineers to specify behavior variation points, contextual information, and non-functional properties for measuring Quality-of-Service (QoS) of the robot. The framework automatically generates the runtime infrastructure for the robot to adapt its behavior and achieve the best QoS according to its current context.