Faisal Alhwikem

Teaching LLMs to Infer Real-World Consequences through Embodied Semantic Grounding

: Large Language Models (LLMs) have recently advanced in real-world commonsense reasoning, including understanding everyday object behaviors and inferring their attributes from text. However, they remain limited in reasoning about the real-world consequences of events, such as how object failures, obstructions, or structural changes affect the surrounding environment-especially without visual or sensorimotor input. Existing works like PIQA and NEWTON evaluate narrow sub-skills, such as whether an object action makes sense and whether object properties can be inferred, providing valuable benchmarks for commonsense and physical reasoning but offering limited evaluation of how events alter environmental functionality and downstream conditions. To address this gap, we propose Embodied Semantic Grounding (ESG), a framework that equips LLMs with consequence-aware text representations. ESG learns a consequence-grounded space by aligning event descriptions with affordance maps-structured representations of how an environment can be used or traversed after an event-capturing how structural changes modify environmental functionality. A Flan-T5-XL model is trained with a contrastive alignment objective to encode event descriptions into this space, for coherent prediction of consequences such as collapses, blockages, and environmental changes. Rather than introducing a new language-model architecture, ESG extends affordance-grounding with consequence-level representations of post-event environmental functionality. We evaluate ESG on a unified benchmark comprising PIQA, NEWTON, LIBERO-derived affordance text, and 2400 synthetic scenario-based tasks. Results show that ESG improves performance over baseline language models across commonsense reasoning and consequence-prediction benchmarks. Under structured affordance-map supervision, ESG improves zero-shot accuracy on PIQA and NEWTON and demonstrates improved performance on synthetic consequence-prediction scenarios designed to evaluate post-event environmental reasoning.

Deep Evolutionary Software Design Using Adaptive Intelligence for Maintainability and Refactoring

Fixed-Time Stable Fault-Tolerant Control of Underactuated Hovercraft via Physics-Informed Neural Adaptation

This study addresses the trajectory tracking control problem for an underactuated hovercraft subject to additive bias and multiplicative loss-of-effectiveness thruster faults under environmental disturbances. In these systems, actuator degradation structurally breaks the differential flatness mapping, driving nominal controllers to generate control actions that induce severe actuator saturation and cause instability. To resolve this challenge, a hierarchical physics-informed neural adaptive control (PINAC) framework is proposed. First, a gated-recurrent-unit physics-informed neural observer (PINO) is designed to isolate thruster faults from exogenous hydrodynamic disturbances. Second, a constrained Safe-TD3 reinforcement learning agent functions as a supervisor, computing an online dilation factor to slow down the mission timeline, thereby reconfiguring the reference trajectory to accommodate degraded actuator boundaries. Third, a low-level non-singular terminal sliding mode (NTSM) controller is implemented as a tracking-guarantee layer. Unlike classical asymptotic schemes where convergence is only achieved as time approaches infinity, or finite-time controllers where the settling time depends on the initial state, the proposed PINAC framework guarantees practical fixed-time stability, ensuring that the settling-time bound is independent of initial conditions. Simulation results demonstrate that the designed controller prevents actuator saturation, provides smooth trajectory adjustment, and reduces tracking errors under severe composite faults.

A Mutation-Guided Safety Assurance Framework for Safety-Critical Cyber-Physical Systems

Safety-critical cyber-physical systems require strict verification methodologies explicitly taken to reference safety properties in their tests. Conventional methods of mutation testing simply lump all mutants together, regardless of their effects on product safety, resulting in ineffective resource allocation and insufficient understanding of the behaviors of crucial safety interest. The novel mutation-guided safety assurance framework presented by this paper is called MuGu and it combines mutation testing and formal safety property enforcement. It uses a hierarchical safety constraint analyzer that classifies mutants (temporally) according to their ability to break temporal safety specifications modeled in Signal Temporal Logic (STL). A graph attention network encodes program semantics and control-flow dependencies to forecast the probability of safety violations, enabling effective prioritization of safety-critical mutants. The proposed safety-conscious mutation delineators use vital areas of the code, such as sensor interfaces, actuator commands, and decision-making code in autonomous systems. Extensive testing on two publicly accessible benchmark sets, namely the Software-artifact Infrastructure Repository (SIR) and Defects4J, shows that MuGu achieves 96.3% safe property coverage and requires 71.8% fewer new tests than traditional methods. The scheme determines 2.4 times as many safety-violating methods as state-of-the-art methods and decreases wasted effort on equivalent mutants by 64.2%. The statistical analysis demonstrates overwhelming improvements across all primary metrics in 12 baseline comparisons p<0.001. MuGu provides a rational basis for applying safety-oriented next-generation mutation testing in autonomous systems.