The application of artificial intelligence (AI) to real-time strategy (RTS) games includes considerable challenges due to the very large state spaces and branching factors, limited decision times, and dynamic adversarial environments involved. To address these challenges, hierarchical task network (HTN) planning has been extended to develop a method denoted as adversarial HTN (AHTN), and this method has achieved favorable performance. However, the HTN description employed cannot express complex relationships among tasks and accommodate the impacts of the environment on tasks. Moreover, AHTN cannot address task failures during plan execution. Therefore, this paper proposes a modified AHTN planning algorithm with failed task repair functionality denoted as AHTN-R. The algorithm extends the HTN description by introducing three additional elements: essential task, phase, and exit condition. If any task fails during plan execution, the AHTN-R algorithm identifies and terminates all affected tasks according to the extended HTN description, and applies a novel task repair strategy based on a prioritized listing of alternative plans to maintain the validity of the previous plan. In the planning process, AHTN-R generates the priorities of alternative plans by sorting all nodes of the game search tree according to their primary features. Finally, empirical results are presented based on a µRTS game, and the performance of AHTN-R is compared to that of AHTN and to the performances of other state-of-the-art search algorithms developed for RTS games.

Multiagent goal recognition is a tough yet important problem in many real time strategy games or simulation systems. Traditional modeling methods either are in great demand of detailed agents’ domain knowledge and training dataset for policy estimation or lack clear definition of action duration. To solve the above problems, we propose a novel Dec-POMDM-T model, combining the classic Dec-POMDP, an observation model for recognizer, joint goal with its termination indicator, and time duration variables for actions with action termination variables. In this paper, a model-free algorithm named cooperative colearning based on Sarsa is used. Considering that Dec-POMDM-T usually encounters multiagent goal recognition problems with different sorts of noises, partially missing data, and unknown action durations, the paper exploits the SIS PF with resampling for inference under the dynamic Bayesian network structure of Dec-POMDM-T. In experiments, a modified predator-prey scenario is adopted to study multiagent joint goal recognition problem, which is the recognition of the joint target shared among cooperative predators. Experiment results show that (a) Dec-POMDM-T works effectively in multiagent goal recognition and adapts well to dynamic changing goals within agent group; (b) Dec-POMDM-T outperforms traditional Dec-MDP-based methods in terms of precision, recall, andF-measure.

Multiagent Reinforcement Learning (MARL) is a promising technique for agents learning effective coordinated policy in Multiagent Systems (MASs). In many MASs, interactions between agents are usually sparse, and then a lot of MARL methods were devised for them. These methods divide learning process into independent learning and joint learning in coordinated states to improve traditional joint state‐action space learning. However, most of those methods identify coordinated states based on assumptions about domain structure (e.g., dependencies) or agent (e.g., prior individual optimal policy and agent homogeneity). Moreover, situations that current methods cannot deal with still exist. In this paper, a modified approach is proposed to learn where and how to coordinate agents’ behaviors in more general MASs with sparse interactions. Our approach introduces sample grouping and a more accurate metric of model difference degree to identify which states of other agents should be considered in coordinated states, without strong additional assumptions. Experimental results show that the proposed approach outperforms its competitors by improving the average agent reward per step and works well in some broader scenarios.

Real-time strategy (RTS) game has proposed many challenges for AI research for its large state spaces, enormous branch factors, limited decision time and dynamic adversarial environment. To tackle above problems, the method called Adversarial Hierarchical Task Network planning (AHTN) has been proposed and achieves favorable performance. However, the HTN description it used cannot express complex relationships among tasks and impacts of environment on tasks. Moreover, the AHTN cannot handle task failures during plan execution. In this paper, we propose a modified AHTN planning algorithm named AHTNR. The algorithm introduces three elements essential task, phase and exit condition to extend the HTN description. To deal with possible task failures, the AHTNR first uses the extended HTN description to identify failed tasks. And then a novel task repair strategy is proposed based on historical information to maintain the validity of previous plan. Finally, empirical results are presented for the μRTS game, comparing AHTNR to the state-of-the-art search algorithms for RTS games.

In the context of robot navigation, game AI, and so on, real‐time search is extensively used to undertake motion planning. Though it satisfies the requirement of quick response to users’ commands and environmental changes, learning real‐time search (LRTS) suffers from the heuristic depressions where agents behave irrationally. There have introduced several effective solutions, such as state abstractions. This paper combines LRTS and encoded quad‐tree abstraction which represent the search space in multiresolutions. When exploring the environments, agents are enabled to locally repair the quad‐tree models and incrementally refine the spatial cognition. By virtue of the idea of state aggregation and heuristic generalization, our EQ LRTS (encoded quad‐tree based LRTS) possesses the ability of quickly escaping from heuristic depressions with less state revisitations. Experiments and analysis show that (a) our encoding principle for quad‐trees is a much more memory‐efficient method than other data structures expressing quad‐trees, (b) EQ LRTS differs a lot in several characteristics from classical PR LRTS which represent the space and refine the paths hierarchically, and (c) EQ LRTS substantially reduces the planning amount and curtails heuristic updates compared with LRTS on uniform cells.

This paper presents the refractive index, absorption coefficient of layered GaSe crystal by means of terahertz time-domain spectroscopy. A unique resonance peak was observed at 0.59 THz (19.6 cm^-1 ), which attributes to the E' phonon mode of epsiv-GaSe.