Khaled Saleh

Traffic incidents remain a critical public safety concern worldwide, with Australia recording 1,300 road fatalities in 2024, which is the highest toll in 12 years. Similarly, the United States reports approximately 6 million crashes annually, raising significant challenges in terms of a fast reponse time and operational management. Traditional response protocols rely on human decision-making, which introduces potential inconsistencies and delays during critical moments when every minute impacts both safety outcomes and network performance. To address this issue, we propose a novel Incident Response Benchmark that uses generative artificial intelligence to automatically generate response plans for incoming traffic incidents. Our approach aims to significantly reduce incident resolution times by suggesting context-appropriate actions such as variable message sign deployment, lane closures, and emergency resource allocation adapted to specific incident characteristics. First, the proposed methodology uses real-world incident reports from the Performance Measurement System (PeMS) as training and evaluation data. We extract historically implemented actions from these reports and compare them against AI-generated response plans that suggest specific actions, such as lane closures, variable message sign announcements, and/or dispatching appropriate emergency resources. Second, model evaluations reveal that advanced generative AI models like GPT-4o and Grok 2 achieve superior alignment with expert solutions, demonstrated by minimized Hamming distances (averaging 2.96-2.98) and low weighted differences (approximately 0.27-0.28). Conversely, while Gemini 1.5 Pro records the lowest count of missed actions, its extremely high number of unnecessary actions (1547 compared to 225 for GPT-4o) indicates an over-triggering strategy that reduces the overall plan efficiency.

Advances in artificial intelligence (AI), including intelligent machines, are opening new possibilities to support teaching and learning in higher education. This research has found a ‘proof of concept’ in the application of machine learning in the assessment of educators’ use of four key microskills, drawn from an internationally established framework. The analysis of teaching videos where these microskills were demonstrated multiple times in front of a green screen or in a space formed the data set. Multiple videos of this nature were recorded to allow for increased analysis and deconstruction of the video components to enable the application of machine learning. The results showed how AI can be used to support the collaborative and reflective practice of educators in a time when online teaching has become the norm. Having achieved a ‘proof of concept’, this research has laid the groundwork to allow for the whole framework of ten microskills to be applied in this way thus adding a new dimension to its use. Providing such critical information that is not currently available in such a systematic and personalised way to educators in the higher education sector can also support the validity of formative assessment practices.

This research showcases the innovative integration of Large Language Models into machine learning workflows for traffic incident management, focusing on the classification of incident severity using accident reports. By leveraging features generated by modern language models alongside conventional data extracted from incident reports, our research demonstrates improvements in the accuracy of severity classification across several machine learning algorithms. Our contributions are threefold. First, we present an extensive comparison of various machine learning models paired with multiple large language models for feature extraction, aiming to identify the optimal combinations for accurate incident severity classification. Second, we contrast traditional feature engineering pipelines with those enhanced by language models, showcasing the superiority of language-based feature engineering in processing unstructured text. Third, our study illustrates how merging baseline features from accident reports with language-based features can improve the severity classification accuracy. This comprehensive approach not only advances the field of incident management but also highlights the cross-domain application potential of our methodology, particularly in contexts requiring the prediction of event outcomes from unstructured textual data or features translated into textual representation. Specifically, our novel methodology was applied to three distinct datasets originating from the United States, the United Kingdom, and Queensland, Australia. This cross-continental application underlines the robustness of our approach, suggesting its potential for widespread adoption in improving incident management processes globally.

Traffic accidents are often inaccurately reported, with incorrect location and disruption duration due to various external factors. This can result in imprecise predictions and inaccurate decision-making in data-driven models. To address these challenges, our study presents a comprehensive framework for traffic disruption segmentation from traffic speed data (obtained from Caltrans Performance Measurements system) in the time-space proximity of reported accidents (from Countrywide Traffic Accident dataset). Furthermore, we evaluate multiple machine learning models on reported, estimated, and manually marked disruption intervals, and demonstrate that our enhanced modelling approach reduces the root mean squared error (RMSE) of traffic accident duration prediction while providing higher similarity with disruptions observed in traffic speed. Our algorithm yields higher disruption detection precision than reported accident timelines. Although using multiple segments offers a slight decrease in the quality of results, it highlights more disruptions. Future research could explore expanding the algorithm’s complexity and applying it to improve traffic incident impact predictions.

Having a profound understanding of the surrounding environment is considered one of the crucial tasks for the reliable operation of future self-driving cars. Light Detection and Ranging (LiDAR) sensor plays a critical role in achieving such understanding due to its capability to perceive the world in 3D. Similar to 2D perception tasks, current state-of-the-art methods in 3D perception tasks rely on deep neural networks (DNNs). However, the performance of 3D perception tasks, specially point-wise semantic segmentation, is not on par with their 2D counterparts. One of the main reasons is the lack of publicly available labelled 3D point cloud datasets (PCDs) from 3D LiDAR sensors. In this work, we are introducing the VoxelScape dataset, a large-scale simulated 3D PCD with 100K annotated point cloud scans. The annotations in the VoxelScape dataset include both point-wise semantic labels and 3D bounding box labels. Additionally, we used a number of baseline approaches to validate the transferability of VoxelScape to real 3D PCD for two challenging 3D perception tasks. The promising results have shown that training DNNs on VoxelScape boosted the performance of the 3D perception tasks on the real PCD. Furthermore, we are also releasing the proposed data generation pipeline for the research community to facilitate realistic simulation of 3D LiDAR point cloud data for different scenarios beyond those covered in our VoxelScape dataset. The VoxelScape dataset and the corresponding LiDAR simulation codes are publicly available at https://voxel-scape.github.io/dataset

This paper describes the co-speech gesture generation system developed by DSI team for the GENEA challenge 2022. The proposed framework features a unique hybrid encoder-decoder architecture based on transformer networks and recurrent neural networks. The proposed framework has been trained using only the official training data split of the challenge and its performance has been evaluated on the testing split. The framework has achieved promising results on both the subjective (specially the human-likeness) and objective evaluation metrics.

Forecasting the trajectory of pedestrians in shared urban traffic environments from non-invasive sensor modalities is still considered one of the challenging problems facing the development of autonomous vehicles (AVs). In the literature, this problem is often tackled using recurrent neural networks (RNNs). Despite the powerful capabilities of RNNs in capturing the temporal dependency in the pedestrians’ motion trajectories, they were argued to be challenged when dealing with longer sequential data. Additionally, whilst the accommodation for contextual information (such as scene semantics and agents interactions) was shown to be effective for robust trajectory prediction, they can also impact the overall real-time performance of prediction system. Thus, in this work, we are introducing a framework based on the transformer networks that were demonstrated recently to be more efficient and outperformed RNNs in many sequential-based tasks. We relied on a fusion of sensor modalities, namely the past positional information, agent interactions information and scene physical semantics information as an input to our framework in order to not only provide a robust trajectory prediction of pedestrians, but also achieve real-time performance for multi-pedestrians’ trajectory prediction. We have evaluated our framework on three real-life datasets of pedestrians in shared urban traffic environments and it has outperformed the compared baseline approaches in both short-term and long-term prediction horizons. For the short-term prediction horizon, our approach has achieved lower scores according to the average displacement error and the root-mean squared error (ADE/RMSE) of predictions over the state-of-the art (SOTA) approach by more than 11 cm and 23 cm, respectively. While for the long-term prediction horizon, our approach has achieved lower ADE and FDE over the SOTA approach by more than 62 cm and 165 cm, respectively. Additionally, our approach has achieved superior real time performance by scoring only 0.025 s (i.e., it can provide 40 individual trajectory predictions per second).