Machine Learning Meets Quantitative Planning: Enabling Self-Adaptation in Autonomous Robots

Abstract

Modern cyber-physical systems (e.g., robotics systems) are typically composed of physical and software components, the characteristics of which are likely to change over time. Assumptions about parts of the system made at design time may not hold at run time, especially when a system is deployed for long periods (e.g., over decades). Self-adaptation is designed to find reconfigurations of systems to handle such run-time inconsistencies. Planners can be used to find and enact optimal reconfigurations in such an evolving context. However, for systems that are highly configurable, such planning becomes intractable due to the size of the adaptation space. To overcome this challenge, in this paper we explore an approach that (a) uses machine learning to find Pareto-optimal configurations without needing to explore every configuration and (b) restricts the search space to such configurations to make planning tractable. We explore this in the context of robot missions that need to consider task timeliness and energy consumption. An independent evaluation shows that our approach results in high-quality adaptation plans in uncertain and adversarial environments.

Automated Online Experiment-Driven Adaptation–Mechanics and Cost Aspects

“…In order to reduce the time cost of a pursuit strategy, CS may have to be pruned offline to reduce the number or range of parameters considered in the pursuit. One way to achieve this is to learn (offline) which parameters have the strongest effect on the system output and use them to form the CS considered in the pursuit [22].…”

“…In the area of adaptive systems, approaches exist that employ online planning to find the best adaptation actions at runtime [22], [31]. A number of algorithms have been employed to this end: Hill climbing has been used to implement a search-based feedback loop [32]; genetic programming and genetic algorithms (including NSGA-II and novelty search) have been advocated as part of the vision of genetic improvement for adaptive software engineering [33] and used in determining optimal configurations [11], [23], [34]- [36]; finally, multi-armed bandits [37] and Bayesian optimization [7], [11] have been employed for online planning in self-adaptive systems.…”

Privacy-Aware UAV Flights through Self-Configuring Motion Planning

“…Autonomous unmanned systems like UAVs are controlled by software with self-adaptive planning methods [18], which decides the planning and subsequent actions in response to uncertainty in the dynamic environment. Although models of software and environment have been leveraged for automated adaptation in robotics software [19]- [21], little has been done to guarantee privacy preservation during UAVs' flights. In this work, we proposed a real-time self-adaptive planner which is responsible for task adaptation (e.g., taking an alternate path) and architectural adaptation (e.g., sensor reconfiguration).…”

Socio-Cyber-Physical Systems: Models, Opportunities, Open Challenges

“…OC5) SCPS self-adaptation-While it makes a lot of sense for SCPS to adapt their configurations (and even their architectures) dynamically to changes in their environment or requirements [28], [29], building self-adaptation capabilities within SCPS is extremely challenging due to the multiple concerns that these systems need to consider. It has taken the research community the best of two decades to explore and advance self-adaptive software systems, and, while this research is relevant to SCPS [30], it needs to be considerably extended before it can address the non-"cyber" aspects of these systems.…”