Structure reasoning is a fundamental capability of large language models (LLMs), enabling them to reason about structured commonsense and answer multi-hop questions. However, existing benchmarks for structure reasoning mainly focus on horizontal and coordinate structures (e.g. graphs), overlooking the hierarchical relationships within them. Hierarchical structure reasoning is crucial for human cognition, particularly in memory organization and problem-solving. It also plays a key role in various real-world tasks, such as information extraction and decision-making. To address this gap, we propose HiBench, the first framework designed to systematically benchmark the hierarchical reasoning capabilities of LLMs from initial structure generation to final proficiency assessment. It encompasses six representative scenarios, covering both fundamental and practical aspects, and consists of 30 tasks with varying hierarchical complexity, totaling 39,519 queries. To evaluate LLMs comprehensively, we develop five capability dimensions that depict different facets of hierarchical structure understanding. Through extensive evaluation of 20 LLMs from 10 model families, we reveal key insights into their capabilities and limitations: 1) existing LLMs show proficiency in basic hierarchical reasoning tasks; 2) they still struggle with more complex structures and implicit hierarchical representations, especially in structural modification and textual reasoning. Based on these findings, we create a small yet well-designed instruction dataset, which enhances LLMs' performance on HiBench by an average of 88.84% (Llama-3.1-8B) and 31.38% (Qwen2.5-7B) across all tasks. The HiBench dataset and toolkit are available at https://github.com/jzzzzh/HiBench to encourage evaluation.

Locomotion has a marked impact on user experience in VR, but currently, common to-go techniques such as steering and teleportation have their limitations. Particularly, steering is prone to cybersickness, while teleportation trades presence for mitigating cybersickness. Inspired by how we manipulate a picture on a mobile phone, we propose illumotion, an optical-illusion-based method that, we believe, can provide an alternative to these two typical techniques. Instead of zooming in a picture by pinching two fingers, we can move forward by “zooming” toward part of the 3D virtual scene with pinched hands. Not only is the proposed technique easy to use, it also seems to minimize cybersickness to some degree.illumotion relies on the manipulation of optics; as such, it requires solving motion parameters in screen space and a model of how we perceive depth. To evaluate it, a comprehensive user study with 66 users was conducted. Results show that, compared with either teleportation, steering or both, illumotion has better performance, presence, usability, user experience and cybersickness alleviation. We believe the result is a clear indication that our novel optically-driven method is a promising candidate for generalized locomotion.

Have you ever picked up a stuffed toy and pretended to play with it in your childhood? We are motivated by the novel use of stuffed toys in enhancing extended reality interaction. A key goal of extended reality is to induce the feeling of presence in its users. Naturally mapped control interface has been shown to enhance presence. The literature also indicates that a high degree of freedom tracking is important to extended reality. Based on these observations, we show that a free-form naturally mapped control interface is well-motivated via a theoretical contextualization. We explore the possibility of building such a controller in the form of stuffed toys. To realize stuffed toys as controllers, a novel soft-pose estimator empowered by cage-based deformation is proposed. It is shown to be effective in tracking the poses and deformations of real soft objects even by training with synthetic data only. Three gameplay prototypes are developed to demonstrate that interactive play can be enabled by the soft-pose estimator. They also form the basis for two user studies that validate the success of tracking stuffed toys with the soft-pose estimator for interactive play.

In recent years, there is a growing phenomenon with children playing mobile devices. Overplaying mobile devices, however, may not lead to a balanced childhood. Lack of physical touch-and-feel stimulus is detrimental to children. Can touch-and-feel stuffed toys be made more appealing to children for a wider variety on their playtime diet? We explore a stuffed toy controller which integrates elements from both digital and make-believe games in an attempt to enrich children interaction experience. A user study was conducted to evaluate the effectiveness of the proposed controller while another children-oriented user study shows that make-believe augmented reality games played by the toy controller can enhance children’s interest in stuffed toys.

In this paper, we are going to present our solid experience in applying constructivism to teach computational thinking and game engineering. Fifty undergraduate students have created nine game based learning applications for a special school in two semesters. This special school caters to children who are severely mentally handicapped, and their day-to-day training consists of repetitive tasks that are designed to help them to express their needs to a certain degree. We will present (1) our timeline and detailed setting of our course which is different from traditional lecture setting. (2) the way to provide situated learning which helps our students to develop better, larger and more linked information. (3) the ways to provide constructivist teaching which helps our students to construct the knowledge of computational thinking and game engineering, so that it can improve their products and have a better interaction with the real world.

Games have grown to be an important part of our society and naturally, its incorporation with education has become a rising trend Together with the strong need for STEM talents, this paper proposed that STEM sandbox game-based learning video games could enhance students' interest and performance in the four disciplines. It is further argued that simulation realism, creativity and interactivity are three important components for this purpose. To promote their development, Planetary Marching Cubes (PMC), a development basis for STEM games has been discussed, along with a game developed with it, HelloPlanet. Students in primary and secondary schools have been invited to play HelloPlanet, and one of the main contributions of this paper is to demonstrate the improvement in both students' interest and performance in STEM. Students and teachers feedback are both very positive, implying that the education sector is fully ready to integrate game-based learning. When comparing with other methods of planet generation, most students picked PMC. This consolidates the argument of the importance of the three components, while also demonstrates the potential of PMC as a development basis for educational sandbox games.

Combined arms are the appropriate combinations to achieve mutually complementary effects in real time strategy (RTS) game. However, it remains the difficulty of planning the sub-unit maneuver, positioning and engagement. Following the research of multi-agents potential fields that provides an effective and efficient way to handle unit maneuver in various environment, we adopted fuzzy measure and integral into potential field and extended the normal additive property to non-additive property in potential field. A new approach for various maneuver planning in RTS game is proposed. The model explains the basic principles that must guide the combined arms in the movement and conduct human-like unit maneuver. We implemented our methodology in a well-known RTS game platform, Warcraft III. In our experiment, fuzzy measure and integral showed the ability to perform various and human-like intelligence in game.

Choquet Integral (CI), which is known as a fuzzy measure-based technique, has been a general aggregation tool for multi-criteria decision making problem. In this paper, we apply Choquet Integral to unit selection problem in Real Time Strategy (RTS) game. In addition, three new fuzzy integrals named Mean based Fuzzy Integral (Me-based FI), Max-based Fuzzy Integral (Ma-based FI), and Order-based Fuzzy Integral (Or-based FI) are developed, which relax the monotonicity requirement of the traditional fuzzy measures and consider different properties of game play. We compare the performance of Choquet Integral and the new proposed ones on this practical application with highly non-monotonic data. Experiments show that the proposed new fuzzy integrals achieved better learning performance and testing result.

Unit formation planning and target of attack is the core of micro management in real time strategy (RTS) game. It is more complicated than macro management, such as building and unit production sequence. It consists of a great quantity of possibility. Multiple targets and the non-additive property of unit formation leads the micro-management remains a problem. Traditional tree searching or A* searching is unable to handle these two properties. They are time consuming in runtime and difficult to manage in the AI programming development as there are too many weightings and each of them will interact with the others. In this paper, we applied potential field, fuzzy measure and integral to solve the micro-management. Potential field is suitable for complicated and various environment with multiple targets. However, it does not consider non-additive property. We integrated it with fuzzy measure and integral to extend simple additive property to non-additive property and provide the ability to handle interaction among different targets.