The agent uses abstract operators that it decomposes into complex combinations of low-level actions. The tank uses its knowledge to select between these activities based on the current situation, just as it selects between different operators. Both of these problems are handled in Soar by allowing abstract high-level operators to be implemented in subgoals where a new state is created that allows low-level operators to be selected in pursuit of the high-level operators.
- Hierarchical task decomposition
- This agent is packaged with the TankSoar environment.
- John Laird
- Soar 8,9