This three step process is repeated for every step of the simulation to allow the strategy to make adjustments as necessary. In Goal Suggestion, the potential goals are filtered into a possible sequence of tasks. In Rule Scoring, possible behaviours from the active rulebook are scored and the best is chosen. In Rule Execution, the selected behaviour is pursued, updating the current task and following the motion.
All three components are currently defined directly by their implementation in code.
As a design principle, a rulebook should contain both high priority rules for safety behaviours, e.g. moving away, and rules that allow it to pursue the goals of the simulation.
For the formal definition of rulebooks and relations between rules for behavior specification, we refer to:
A. Censi et al., "Liability, Ethics, and Culture-Aware Behavior Specification using Rulebooks,"
2019 International Conference on Robotics and Automation (ICRA), Montreal, QC, Canada, 2019, pp. 8536-8542, doi: 10.1109/ICRA.2019.8794364.
Specific implemented Strategies
Our currently implemented strategies for the purpose of radiation therapy include:
Fixed Order
The fixed order strategy resembles the conventional standard procedure for robotic radiation therapy. This strategy simply follows the pre-computed list of beams in its ordering. If the motion interferes, the radiation delivery is simply paused.
Eager Movement
Our eager movement strategy is rather simple and contains rules that move beam delivery further away from or nearer to the patient based on the currently active goal. Furthermore, there is a high priority rule to move the equipment further away if a collision due to the patient's respiratory motion is expected in the near future.
Best Verified
The best verified strategy is an improved version of the fixed order strategy and verifies possible beam delivery based on an online model.
It similarly tries to follow the pre-computed list, but diverges whenever necessary to reduce treatment pauses as much as possible.
This is further refined with a pre-selection of the checked beams, where high-priority beams are checked first and are thusly guaranteed to be within the time window for the decision. Primarily, we try to stay with the currently radiating beam to reduce configuration changes.
Furthermore, there is also the rule to evade collisions for additional safety.
More information on this strategy is provided in our paper.
