This agent demonstrates how to solve the water jug problem using reinforcement learning. It is a modification of the Simple Water Jug Agent. Two main changes are made: templates are used to generate the full space of possible moves that the agent can perform and rewards are set based on the problem solution.
agent incorporates both operator subgoaling/means ends analysis with reinforcement learning. All search control knowledge (operator evaluation rules) are removed from blocks-world-operator-subgoaling and instead there are RL rules supplemented with rules to compute reward, both in the top state and the substate. Implemented for four blocks.
This agent contains a version of blocks-world that involves one level of problem spaces and look-ahead but with two important extensions. It demonstrates how RL-rules can be learned by chunking and then updated in the future. The advantage over simple lookahead is that it doesn't lock on to the one path found during look-ahead after chunking. It will still do some exploration.
An agent that performs graph search using the selection-astar default rules. This approach can be modified to use any of the different selection approaches.
The basic idea behind this agent is that there is a mission to go from place to place (in the mission structure) using go-to-location for each go-to-location use an iterative form of A* search to find the minimal path between each place iteratively select go-to-waypoint to move through the graph
Key data structures (initialized...
An agent that tests four rl-rule sequence cases to behaviorally test the Soar reinforcement learning update mechanism. Learning/discount-rate are set to known values and other parameters can be tweaked. Some of the expected behavior is documented in the README file included.