Bootstrapping Monte Carlo Tree Search with an Imperfect Heuristic
This work addresses planning efficiency in non-adversarial settings like large-state MDPs, offering an incremental improvement over existing UCT methods.
The paper tackled the problem of improving the Upper Confidence Bound applied in Trees (UCT) algorithm for planning in large-state Markov Decision Processes by incorporating an imperfect heuristic policy, resulting in the UCT-Aux algorithm that outperformed UCT and its variants in benchmark experiments.
We consider the problem of using a heuristic policy to improve the value approximation by the Upper Confidence Bound applied in Trees (UCT) algorithm in non-adversarial settings such as planning with large-state space Markov Decision Processes. Current improvements to UCT focus on either changing the action selection formula at the internal nodes or the rollout policy at the leaf nodes of the search tree. In this work, we propose to add an auxiliary arm to each of the internal nodes, and always use the heuristic policy to roll out simulations at the auxiliary arms. The method aims to get fast convergence to optimal values at states where the heuristic policy is optimal, while retaining similar approximation as the original UCT in other states. We show that bootstrapping with the proposed method in the new algorithm, UCT-Aux, performs better compared to the original UCT algorithm and its variants in two benchmark experiment settings. We also examine conditions under which UCT-Aux works well.