Multimodal Large Language Models (MLLMs) harness comprehensive knowledge spanning text, images, and audio to adeptly tackle complex problems. This study explores the ability of MLLMs in visually solving the Traveling Salesman Problem (TSP) and Multiple Traveling Salesman Problem (mTSP) using images that portray point distributions on a two-dimensional plane. We introduce a novel approach employing multiple specialized agents within the MLLM framework, each dedicated to optimizing solutions for these combinatorial challenges. We benchmarked our multi-agent model solutions against the Google OR tools, which served as the baseline for comparison. The results demonstrated that both multi-agent models—Multi-Agent 1, which includes the initializer, critic, and scorer agents, and Multi-Agent 2, which comprises only the initializer and critic agents—significantly improved the solution quality for TSP and mTSP problems. Multi-Agent 1 excelled in environments requiring detailed route refinement and evaluation, providing a robust framework for sophisticated optimizations. In contrast, Multi-Agent 2, focusing on iterative refinements by the initializer and critic, proved effective for rapid decision-making scenarios. These experiments yield promising outcomes, showcasing the robust visual reasoning capabilities of MLLMs in addressing diverse combinatorial problems. The findings underscore the potential of MLLMs as powerful tools in computational optimization, offering insights that could inspire further advancements in this promising field.

Currently, a growing number of cities are adopting the Median-U-Turn (MUT) intersection design to enhance road capacity and traffic efficiency. The critical question in selecting the right intersection design is how significantly the implementation of MUT can enhance intersection performance, focusing on three key aspects: intersection efficiency, capacity, and the environmental impact of the design. To address this question, an evaluation of operational performance under various prevailing conditions (roadway and control) was conducted using VISSIM, a microscopic simulation platform. This evaluation involved five scenarios: conventional intersections (with increased cycle length, grade separation with a signalized at-grade intersection, grade separation with a roundabout), MUT, and signalized crossover MUT at a dense urban arterial intersection in Amman, Jordan's capital. The performance was compared using several metrics: average control delay, number of stops, average travel speed and time, average stopped delay, Carbon Monoxide (CO) emissions, fuel consumption, and vehicle safety. The Criteria Importance Through Intercriteria Correlation (CRITIC) technique was subsequently used to select the optimal design. The findings indicate that the existing intersection configuration is the least effective, while the MUT with signals at the crossing U-turn points is the most efficient solution.

: Multimodal Large Language Models (MLLMs) harness comprehensive knowledge spanning text, images, and audio to adeptly tackle complex problems, including zero-shot in-context learning scenarios. This study explores the ability of MLLMs in visually solving the Traveling Salesman Problem (TSP) and Multiple Traveling Salesman Problem (mTSP) using images that portray point distributions on a two-dimensional plane. We introduce a novel approach employing multiple specialized agents within the MLLM framework, each dedicated to optimizing solutions for these combinatorial challenges. Our experimental investigation includes rigorous evaluations across zero-shot settings and introduces innovative multi-agent zero-shot in-context scenarios. The re-sults demonstrated that both multi-agent models—Multi-Agent 1, which includes the Initializer, Critic, and Scorer agents, and Multi-Agent 2, which comprises only the Initializer and Critic agents—significantly improved solution quality for TSP and mTSP problems. Multi-Agent 1 ex-celled in environments requiring detailed route refinement and evaluation, providing a robust framework for sophisticated optimizations. In contrast, Multi-Agent 2, focusing on iterative re-finements by the Initializer and Critic, proved effective for rapid decision-making scenarios. These experiments yield promising outcomes, showcasing the robust visual reasoning capabilities of MLLMs in addressing diverse combinatorial problems. The findings underscore the potential of MLLMs as powerful tools in computational optimization, offering insights that could inspire further advancements in this promising field.

Young drivers remain disproportionately represented in crash and fatality statistics worldwide, with inexperience and skills deficiencies identified as critical factors. Current state of practices, such as Graduated Licensing Systems (GLS), have reduced crash risks but may have some limitations in preparing drivers for independent operation. In response, technological interventions ranging from simulation-based studies and field-based studies to emerging multimodal large language model (MLLM) applications have been developed to provide both retrospective feedback and real-time advice. This systematic review analysed 46 studies retrieved from six databases (Scopus, IEEE Xplore, ScienceDirect, Web of Science, TRID, and Taylor & Francis) published between 2006 and 2025. The types of technology-based intervention systems designed to improve novice driver performance, the driving skills targeted, the modalities of data collection, the effectiveness of different intervention strategies were analysed. This review found that current systems in literature (field-based, simulationbased, and emerging AI-based) target a limited range of skills, with most focusing on low-order driving skills such as vehicle control, vehicle manoeuvring, and road rules. Field studies relied mainly on kinematic data, occasionally supplemented by manually annotated videos, and emphasised retrospective feedback, which rarely produced significant behavioural changes unless combined with incentives or parental involvement. Real-time interventions were largely restricted to simple auditory or visual reactive alerts. Simulation studies utilised richer, multimodal data and supported diverse interventions, achieving improvements in skills such as vehicle control, observation, and hazard perception, although the effects remained confined to 2 virtual settings. Emerging AI-driven approaches using LLMs and MLLMs show promise but remain in their infancy. Research gaps and future research directions needed to develop a coaching system for novice drivers were identified. Persistent challenges include reliance on single-modality data, lack of real-world validation, centralized architectures, and the absence of human-centric, prioritized real-time advice. This review highlights the potential of lightweight, edge-deployable MLLMs, multimodal integration, adaptive prioritization frameworks, lack of personalised and adaptive coaching systems, and limited number of targeted skills to overcome these limitations. Collectively, the findings underscore the need for comprehensive, skill-based, and personalised coaching systems that can further prepare novice drivers for the complexities of independent driving.