1. Develop a novel planning algorithm that efficiently achieves a new, previously unknown objective given the current environment model of the system. The planning algorithm should account for the fact that the environment model may change over time.
  2. Develop a novel exploration strategy for RL that automatically and efficiently updates the environmental model, by selecting actions that explore parts of the environment that the system is not yet familiar with.
  3. Develop a novel framework for task decomposition that automatically creates and evaluates tasks, discarding tasks that are not deemed useful. Each task has its own associated decision strategy.
  4. Evaluate the novel planning and RL algorithms in two realistic scenarios: active network management for electrical distribution networks, and microgrid management. Apart from using these scenarios for evaluation, the project also aims at improving on the state-of-the-art in these two applications.