Self-Adaptation in Autonomous Robots with Machine Learning
When Machine Learning Marries Quantitative Planning
This research summary is just one of many that are distributed weekly on the AI scholar newsletter. To start receiving the weekly newsletter, sign up here.
Contemporary software-intensive systems integrate components whose behavior is expected to change over time. Think third-party web services whose accessibility or performance varies due to implementation changes or controllers in cyber-physical systems such as robotics where reliability gradually decrease due to hardware wear and tear.
Due to the changes in assumptions made at design time for such systems, software might not be able to operate as expected. Furthermore, the issue may affect the expected behavior of self-adaptive components which degrade capacity.
Machine Learning Meets Quantitative Planning
Researchers have proposed an integrated learning and quantitative planning approach whose main goal to enable self-adaptation in highly-configurable systems that operate in dynamic and uncertain environments such as robotic systems. The technique uses configuration changes as the primary mechanism to enact adaptation.
The new approach is innovative in that it applies machine learning to discover Pareto-optimal configurations without the need to explore all configurations and applies restrictions of the search space to the specific settings for controllable planning. This way it can incorporate both learning and quantitative planning to allow run time self-adaptations. Moreover, the approach facilitates the integration of information from several models in quantitative planning. Specifically, the researchers explore robot operations that need to consider timeliness and energy consumption. An independent evaluation has demonstrated that the proposed method results in high-quality adaptation procedures in uncertain and dynamic environments.
Potential Uses and Effects
As a new technique that effectively enables integrated learning and quantitative planning approach to implement robotic adaptation, the method can be used with many other cyber-physical systems. Further, it can be extended to account for model updates at run time as an online interface.
Thanks for reading. Please comment, share and don’t forget to subscribe! Also, follow me on Twitter and LinkedIn. Remember to 👏 if you enjoyed this article. Cheers!
