This issue presents a collection of research that explores intelligence as an adaptive, interactive, and autonomous capability. The articles in this issue span generative environment modeling, trustworthy distributed systems, human-computer interaction, embodied robotics, optimization, and learning under dynamic conditions. Despite their diverse applications, these contributions are united by a common objective: developing systems that can perceive complex environments, make informed decisions, and collaborate effectively with humans and other agents. Together, they illustrate the growing convergence of artificial intelligence, autonomous systems, and agent-based computing in addressing real-world challenges.
Published
June 2026
Contributions
10 articles
Access
Open access

Explore the research published in this issue of ijaias.
Prashanth Chevva
In this paper, we present SmartSnake, a flexible software environment based on Dionysos.jl for the intelligent synthesis of abstract-based control of complex dynamical systems. The platform offers tools that allow researchers and practitioners to design custom abstraction algorithms, which support partial state-space coverage and state-dependent controllers. The modular architecture of SmartSnake allows for an easy integration of machine learning and artificial intelligence to help develop future meta-solvers that will adapt intelligently to the structure of problems and overcome the curse of dimensionality as well. The solvers that can be utilized in SmartSnake are evaluated with respect to how practical they are and their computational efficiency for control synthesis scenarios. We will also present ongoing work on improvement of the platform with data-driven methods and heuristic symbolic models. SmartSnake provides a flexible framework for both reinforcement learning approaches and classical optimization of controllers.
Ujjwal Yadav
Trust is a core concept in secure and cooperative distributed systems. However, the dynamic, context-dependent nature of trust poses challenges to its formalization and enforcement. In this paper, we offer a framework for trust management in a modal logic-based setting. In particular, we offer a rigorous method to model belief, knowledge, and delegation among agents. In multi-agent systems, we study how modal operators can represent fundamental aspects of trust, such as permissions, obligations, and authority propagation. Our framework improves the specification and verification of trust-based decisions, enabling reasoning through modal logic to formalize trust relationships. The system were proposed can induce both monotonic and non-monotonic inferences, thus providing flexibility in inducing or revoking trust. The framework is demonstrated to be wide-ranging, robust, and useful in policy specification, access control, and automated trust negotiation. The proposed work promotes transparent, verifiable, and intelligent trust management mechanisms in distributed and decentralized systems.
Eeshwar Pasula
As AI agents are increasingly being integrated into work, little is known about the systematic knowledge among workers. We conduct a large audit of 1,500 workers across 104 US occupations, examining the preferred AI involvement in 844 occupational tasks. Our main contribution is the Human Agency Scale (HAS), which comprises 5 levels of human involvement between fully automated (H1) and essential human agency (H5). HAS seeks to circumvent the automation debate. The findings show that there is a 46.1% positive tendency to auto-automate tasks. According to research, this was largely to free up time for higher-value work (69.4%). Moreover, resistance has come from the creative sector as well. A significant mismatch is indicated by the comparison between what workers want and what 52 AI specialists think. We illustrate these mismatches as a desire-capability landscape consisting of four zones. The zones are Automation “Green Light”, Automation “Red Light”, R&D Opportunity, and Low Priority. Currently, investment is mostly concentrated in low-priority zones with unmet labour needs. H3 (Equal partnership) occupies 45.2% of the jobs, indicating an upcoming phase of collaboration. Interpersonal skills are becoming increasingly important. The implementation of AI technology and focused research should be guided by this worker-centred framework.
Monisha Rengaraj
In this paper, an approach to global path planning for unmanned surface vehicles is recast as a problem closer to embedded systems rather than that of artificial intelligence (AI) by the integration of a chemotaxis operator with an A*‑like heuristic combined with bacterial foraging. The resulting global path planner enables deterministic low-latency navigation on resource-constrained controllers. A model of the grid-partitioned environment makes bounded-memory representations feasible, while its fixed-cost neighbour evaluation assists in timing budgets used in the common case. Disruptive tumbling is avoided via a continuity checker that only invokes when a tumble will break path connectivity. Paths are represented as variable-length sequences that are mapped to compact data structures for the on-board execution. The algorithm parameters are treated as design-space knobs that co-optimise the convergence as well as the trajectory length and the iteration counts with due regard to compute and memory constraints. According to the Morris method, reproduction and elimination–dispersal are the most influential factors embedded in the scheduling. Simulation studies with five grid sizes show this approach produces shorter paths when compared with GA and ACO with fewer iterations. Also, with growing map sizes, the number of iterations becomes predictable for embedded deployments. The architecture thus obtained is a ready-to-implement template for a USV controller: a FIFO-friendly neighbourhood scan, a continuity-aware tumbling and event-triggered A* repairs that conform to standard microcontroller or SoC constraints.
Guruprasath Sankaran
The emergence of generative search engines (GES), including ChatGPT Search, Perplexity AI, and Google SGE, has transformed information retrieval by generating synthesized answers rather than ranked hyperlinks. Consequently, traditional Search Engine Optimization (SEO) is becoming less effective due to declining organic click-through rates and the growth of zero-click searches. This paper introduces Generative Engine Optimization (GEO), a systematic framework for improving content visibility within AI-generated responses. We propose a three-layer semantic visibility model consisting of Semantic Anchoring (clear topical organization), Context Triggering (semantic coverage through synonyms and domain-specific terminology), and Pragmatic Recomposition (modular, extractable content using FAQs, lists, and standalone facts). The framework is implemented using static HTML, Schema.org JSON-LD markup, and a semantic mesh architecture. GEO is evaluated through two real-world case studies: a commercial course review page (SOYA). The study investigates how semantic structuring influences citation visibility, the contribution of each semantic layer, and when GEO outperforms authority-based SEO signals. Results show citation rates increasing from 0% to 77.1% across ChatGPT and Perplexity despite poor traditional search rankings, while SEO rankings alone failed to produce generative citations. Five quantitative metrics are introduced to assess GEO readiness and guide optimization for AI-driven search.
Elvis Mondal
Modular reconfigurable robots have emerged as a promising paradigm for adaptive locomotion and multifunctional structural systems. However, their widespread adoption is constrained by complex mechanical designs, high manufacturing costs, and limited accessibility. This paper presents the Gluss system, a modular biomimetic robotic architecture based on variable-geometry trusses integrated with a 3D-printable spherical turret joint that enables both locomotion and load-bearing functionality. The proposed system continuously regulates actuator lengths and structural geometry through a dynamic conformation feedback mechanism to maintain stability during adaptive crawling and oozing locomotion across diverse terrains. A theoretical framework is developed by deriving the optimal actuator length ratio from geometric constraints, demonstrating that the achievable limit converges to the golden ratio (φ = 1.618). The hardware implementation employs commercially available linear actuators, Arduino-based control electronics, and additive manufacturing, resulting in an accessible, low-cost, and reproducible platform for modular robotics research. Experimental validation was conducted using the 3TetGlussBot and 5TetGlussBot prototypes, which achieved a maximum locomotion speed of 27~cm/min while successfully traversing multiple terrain types. These results demonstrate the feasibility of dynamic conformation control for scalable modular robotic systems with potential applications in search and rescue, adaptive infrastructure, and programmable robotic materials.
Elvis Mondal
Despite significant advances in the area of inclusive game development, digital games are still mainly visual experiences that create important accessibility barriers for blind and visually impaired players. This study examines game accessibility within the framework of human-computer interaction (HCI) and proposes a structured set of Universally Accessible Game Design (UAGD) principles. By reviewing all the literature and current industry practice, the study identified visual dependency, navigation complexity, cognitive load, technical implementation limitations, and lack of accessibility standards as the key challenges. In order to overcome these limitations, we propose a user-centred development framework aimed at embedding accessibility into the game-development life cycle and multimodal interaction patterns by bringing together spatial audio, haptics, cognitive support mechanisms, and adaptive navigation techniques. This study further explores the emerging technologies in AI, haptic devices, and AI-augmented echolocation systems, which can improve non-visual interaction and personalization. Instead of suggesting the next game engine or implementation, the paper offers some useful design patterns and development workflows. Moreover, several research directions that have the potential to help researchers and developers create more inclusive digital games are suggested. The framework proposed aims to mitigate cognitive load while enhancing spatiotemporal awareness and equity within interactive experience development settings.
Guruprasath Sankaran
Automatic License Plate Recognition (ALPR) systems deployed in urban environments process continuous, high-velocity data streams under inherently non-stationary conditions. However, most machine learning approaches assume stationary data and rely on offline evaluation, creating a gap between experimental validation and operational reality. This paper addresses this gap through a controlled comparison of static and adaptive learning strategies for rare-event detection in simulated ALPR streams. We develop a simulation testbed that independently controls operational scale, rare-event prevalence, and temporal data drift. Six supervised models—three static (Logistic Regression, Random Forest, HistGradientBoosting) and three adaptive incremental learners (Hoeffding Tree, Adaptive Random Forest, Leveraging Bagging)—are evaluated using a prequential test-then-train protocol across class prevalence levels of 1–5% with controlled drift. Results show that adaptive models achieve F1-scores (0.927–0.930) comparable to static models (0.921–0.928), with overlapping 95% confidence intervals, indicating no significant accuracy advantage under stable conditions. Computational analysis reveals important trade-offs: Hoeffding Tree processes 11,920 instances per second—747× faster than Random Forest—with only a 0.6% F1 reduction, while Adaptive Random Forest offers a balanced compromise between accuracy and throughput. Under simulated distribution shift, adaptive models generate substantially fewer false positives than static models, demonstrating greater robustness. These findings provide practical guidance for selecting learning architectures for privacy-preserving, real-time ALPR systems operating under evolving data distributions.
Jitendra Gupta
Adaptive activation functions have emerged as a promising approach for improving neural network expressiveness without increasing architectural complexity. This paper investigates whether trainable nonlinearities can enhance regression performance in compact feedforward networks operating under limited computational budgets. We propose a modified Extreme Learning Machine (ELM) framework in which activation functions are parameterized and learned through two adaptation strategies: layer-wide parameter sharing and neuron-specific parameterization. Four activation functions, quadratic, PReLU, sigmoid, and swish, are evaluated in both their standard and adaptive forms across ten diverse regression benchmark datasets. Experimental results demonstrate that adaptive parameterization consistently improves predictive performance over fixed activation functions while introducing only a modest increase in computational cost. Neuron-specific adaptation achieves the lowest root mean squared error (RMSE) in the majority of datasets, whereas layer-wide adaptation offers a favorable trade-off between accuracy and parameter efficiency. Among the investigated nonlinearities, adaptive quadratic functions yield the largest improvements, with average error reductions exceeding 18% relative to their standard counterparts. Statistical analyses using one-way ANOVA and Tukey's post-hoc tests confirm that the observed performance gains are significant across most datasets. These findings indicate that functional plasticity through trainable activation parameters provides an effective alternative to increasing network depth or width, enabling more accurate and computationally efficient regression models for resource-constrained applications.
Jitendra Gupta
This paper proposes a novel architecture for solving high-dimensional visual control tasks without relying on recurrent neural networks (RNNs) or stacked image frames as input. Instead of directly learning policies from raw observations, the proposed framework first constructs a generative model that captures the underlying dynamic processes of the environment. This generative model is then utilized to train the reinforcement learning agent on a compact and informative latent representation of the environment state. By learning a structured latent space, the architecture effectively encodes temporal dependencies and environmental dynamics while reconstructing the complete sequence of observed states. Consequently, the learned representations provide a more meaningful description of the environment than raw visual inputs alone. Furthermore, the probabilistic nature of the generative model enables the framework to capture transition uncertainty, allowing the agent to make more robust decisions under stochastic conditions and improving learning stability. The proposed approach accelerates policy convergence by reducing the complexity of high-dimensional observations while preserving essential dynamic information. The agent operates without any prior knowledge of the environment and learns directly through interaction in a continuous action space using an on-policy actor-critic reinforcement learning algorithm. Experimental results demonstrate that the proposed architecture achieves efficient representation learning, faster convergence, and improved policy performance across challenging visual control environments.
Read the articles online or download individual PDFs from this issue.