Teaching People by Justifying Tree Search Decisions: An Empirical Study in Curling

In this research note we show that a simple justification system can be used to teach humans non-trivial strategies of the Olympic sport of curling. This is achieved by justifying the decisions of Kernel Regression UCT (KR-UCT), a tree search algorithm that derives curling strategies by playing the game with itself. Given an action returned by KR-UCT and the expected outcome of that action, we use a decision tree to produce a counterfactual justification of KR-UCT’s decision. The system samples other possible outcomes and selects for presentation the outcomes that are most similar to the expected outcome in terms of visual features and most different in terms of expected end-game value. A user study with 122 people shows that the participants who had access to the justifications produced by our system achieved much higher scores in a curling test than those who only observed the decision made by KR-UCT and those with access to the justifications of a baseline system. This is, to the best of our knowledge, the first work showing that a justification system is able to teach humans non-trivial strategies learned by an algorithm operating in self play.

Monte caro tree search algorithm for the receiving containers intelligently problem among container shipping terminals

With the development of economic globalization, shipping exchanges between countries are rapidly increasing and the container throughput of shipping ports has increased quickly, and the operation of most seaport terminals in China has reached a bottleneck which puts forward new requirements for the efficiency of container shipping terminal operations. Therefore, locating for export containers intelligently is of great significance for the development of terminals. This paper focuses on receiving containers intelligently, and establishes the export container locating model based on the principle of actual shipping terminal operations. The Monte Carlo Tree Search algorithm for export container locating problem was proposed and constructed. After examples, the algorithm can effectively solve the problem that meet the constraints, which further proves the practicability of the algorithm and the correctness of the model. The research shows that the method of locating for export container based on Monte Carlo Tree Search algorithm can effectively solve the problem and maintain the green, energy-saving and sustainable development of the shipping terminal. The ideas and methods have certain academic value and reference significance for other NP-Hard problems.

Gesture commands for controlling high-level UAV behavior

Directing groups of unmanned air vehicles (UAVs) is a task that typically requires the full attention of several operators. This can be prohibitive in situations where an operator must pay attention to their surroundings. In this paper we present a gesture device that assists operators in commanding UAVs in focus-constrained environments. The operator influences the UAVs’ behavior by using intuitive hand gesture movements. Gestures are captured using an accelerometer and gyroscope and then classified using a logistic regression model. Ten gestures were chosen to provide behaviors for a group of fixed-wing UAVs. These behaviors specified various searching, following, and tracking patterns that could be used in a dynamic environment. A novel variant of the Monte Carlo Tree Search algorithm was developed to autonomously plan the paths of the cooperating UAVs. These autonomy algorithms were executed when their corresponding gesture was recognized by the gesture device. The gesture device was trained to classify the ten gestures and accurately identified them 95% of the time. Each of the behaviors associated with the gestures was tested in hardware-in-the-loop simulations and the ability to dynamically switch between them was demonstrated. The results show that the system can be used as a natural interface to assist an operator in directing a fleet of UAVs. Article highlights A gesture device was created that enables operators to command a group of UAVs in focus-constrained environments. Each gesture triggers high-level commands that direct a UAV group to execute complex behaviors. Software simulations and hardware-in-the-loop testing shows the device is effective in directing UAV groups.

Incorporation of Potential Fields and Motion Primitives for the Collision Avoidance of Unmanned Aircraft

Collision avoidance (CA) using the artificial potential field (APF) usually faces several known issues such as local minima and dynamically infeasible problems, so unmanned aerial vehicles’ (UAVs) paths planned based on the APF are safe only in a certain environment. This research proposes a CA approach that combines the APF and motion primitives (MPs) to tackle the known problems associated with the APF. Since MPs solve for a locally optimal trajectory with respect to allocated time, the trajectory obtained by the MPs is verified as dynamically feasible. When a collision checker based on the k-d tree search algorithm detects collision risk on extracted sample points from the planned trajectory, generating re-planned path candidates to avoid obstacles is performed. After rejecting unsafe route candidates, one applies the APF to select the best route among the remaining safe-path candidates. To validate the proposed approach, we simulated two meaningful scenario cases—the presence of static obstacles situation with local minima and dynamic environments with multiple UAVs present. The simulation results show that the proposed approach provides smooth, efficient, and dynamically feasible pathing compared to the APF.

Enhanced Reinforcement Learning Method Combining One-Hot Encoding-Based Vectors for CNN-Based Alternative High-Level Decisions

Gomoku is a two-player board game that originated in ancient China. There are various cases of developing Gomoku using artificial intelligence, such as a genetic algorithm and a tree search algorithm. Alpha-Gomoku, Gomoku AI built with Alpha-Go’s algorithm, defines all possible situations in the Gomoku board using Monte-Carlo tree search (MCTS), and minimizes the probability of learning other correct answers in the duplicated Gomoku board situation. However, in the tree search algorithm, the accuracy drops, because the classification criteria are manually set. In this paper, we propose an improved reinforcement learning-based high-level decision approach using convolutional neural networks (CNN). The proposed algorithm expresses each state as One-Hot Encoding based vectors and determines the state of the Gomoku board by combining the similar state of One-Hot Encoding based vectors. Thus, in a case where a stone that is determined by CNN has already been placed or cannot be placed, we suggest a method for selecting an alternative. We verify the proposed method of Gomoku AI in GuPyEngine, a Python-based 3D simulation platform.