Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Daniele De Martini has a total of 119 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • PriorEye: Geospatial Visual Priors for End-to-End Autonomous Driving

    by Kyuhwan Yeon, Benjamin Ramtoula, Daniele De Martini

  • Open Access

    Fantastic Features and Where to Find Them: A Probing Method to combine Features from Multiple Foundation Models

    by Benjamin Ramtoula, Pierre-Yves Lajoie, Paul Newman, Daniele De Martini

    Foundation models (FMs) trained with different objectives and data learn diverse representations, making some more effective than others for specific downstream tasks. Existing adaptation strategies, such as parameter-efficient fine-tuning, focus on individual models and do not exploit the complementary strengths across models. Probing methods offer a promising alternative by extracting information from frozen models, but current techniques do not scale well with large feature sets and often rely on dataset-specific hyperparameter tuning. We propose Combined backBones (ComBo), a simple and scalable probing-based adapter that effectively integrates features from multiple models and layers. ComBo compresses activations from layers of one or more FMs into compact token-wise representations and processes them with a lightweight transformer for task-specific prediction. Crucially, ComBo does not require dataset-specific tuning or backpropagation through the backbone models. However, not all models are equally relevant for all tasks. To address this, we introduce a mechanism that leverages ComBo's joint multi-backbone probing to efficiently evaluate each backbone's task-relevance, enabling both practical model comparison and improved performance through selective adaptation. On the 19 tasks of the VTAB-1k benchmark, ComBo outperforms previous probing methods, matches or surpasses more expensive alternatives, such as distillation-based model merging, and enables efficient probing of tuned models. Our results demonstrate that ComBo offers a practical and general-purpose framework for combining diverse representations from multiple FMs.

  • Open Access

    Robot-relay: building-wide, calibration-less visual servoing with learned sensor handover networks

    by Luke Robinson, Matthew Gadd, Paul Newman, Daniele De Martini

    This paper proposes a novel system to conduct visual servoing of a mobile robot using multiple uncalibrated, wall-mounted cameras. Specifically, we utilise a constellation of such sensors to cover a wide area by allowing robot control to be passed between cameras in regions where their fields of view overlap. This method, in conjunction with the fact that all computing is also executed offboard, allows for simpler and cheaper robots to be deployed in controlled and finite spaces. Our method simplifies the natural installation cycle of a newly deployed camera network, eliminating the need for explicit camera positioning or orientation, both globally (relative to a building plan) and locally (among viewpoints). Our system memorises pixel-wise topological connections between viewpoints by leveraging natural human exploration of the environment. We detect graph edges through simultaneous detections of the same person across different cameras, allowing us to automatically construct an evolving graph that represents overlapping fields of view within the camera network. In combination with a hybrid-A*-based planner, our approach allows efficient planning and control of robots across a wide area by traversing cameras between areas of overlap. We validate our approach through autonomous traversals in a productive office environment, using a network of six cameras, and compare our performance against both human teleoperation and a traditional Simultaneous Localisation and Mapping (SLAM) approach.


  • 3D Foundation Model-Based Loop Closing for Decentralized Collaborative SLAM

    by Pierre-Yves Lajoie, Benjamin Ramtoula, Daniele De Martini, Giovanni Beltrame

    Decentralized Collaborative Simultaneous Localization and Mapping (C-SLAM) techniques often struggle to identify map overlaps due to significant viewpoint variations among robots. Motivated by recent advancements in 3D foundation models, which can register images despite large viewpoint differences, we propose a robust loop closing approach that leverages these models to establish inter-robot measurements. In contrast to resource-intensive methods requiring full 3D reconstruction within a centralized map, our approach integrates foundation models into existing SLAM pipelines, yielding scalable and robust multi-robot mapping. Our contributions include: (1) integrating 3D foundation models to reliably estimate relative poses from monocular image pairs within decentralized C-SLAM; (2) introducing robust outlier mitigation techniques critical to the use of these relative poses; and (3) developing specialized pose graph optimization formulations that efficiently resolve scale ambiguities. We evaluate our method against state-of-the-art approaches, demonstrating improvements in localization and mapping accuracy, alongside significant gains in computational and memory efficiency. These results highlight the potential of our approach for deployment in large-scale multi-robot scenarios.


  • Select2Plan: Training-Free ICL-Based Planning Through VQA and Memory Retrieval

    by Davide Buoso, Luke Robinson, Giuseppe Averta, Philip Torr, Tim Franzmeyer, Daniele De Martini

    We introduce Select2Plan (S2P), a novel training-free framework for high-level robot planning that leverages off-the-shelf VLMs for autonomous navigation. Unlike most learning-based approaches that require extensive task-specific training and large-scale data collection, S2P overcomes the need for fine-tuning by adapting inputs to align with the VLM's pretraining data. Our method achieves this through a combination of structured Visual Question Answering (VQA) to ground action selection on the image, and In-Context Learning (ICL) to exploit knowledge drawn from relevant examples from a memory bank of (visually) annotated data, which can include diverse, in-the-wild sources. We demonstrate S2P flexibility by evaluating it in both First-Person View (FPV) and Third-Person View (TPV) navigation. S2P improves the performance of a baseline VLM by 40% in TPV and surpasses end-to-end trained models by approximately 24% in FPV when tasked with navigating towards unseen objects in novel scenes. These results highlight the adaptability, simplicity, and effectiveness of our training-free approach, demonstrating that the use of pre-trained VLMs with structured memory retrieval enables robust high-level robot planning without costly task-specific training. Our experiments also show that retrieving samples from heterogeneous data sources, including online videos of different robots or humans walking, is highly beneficial for navigation. Notably, our method effectively generalizes to novel scenarios, requiring only a handful of demonstrations. Project Page: lambdavi.github.io/select2plan


  • AutoInspect: Toward Long-Term Autonomous Inspection and Monitoring

    by Michal Staniaszek, Tobit Flatscher, Joseph Rowell, Hanlin Niu, Wenxing Liu, Yang You, Matthew Gadd, Matías Mattamala, Alex Schutz, Daniele de Martini, et al.

    This paper presents a technical overview of AutoInspect, a ROS-based software system for robust and extensible mission-level autonomy. Over the past three years we have deployed AutoInspect on multiple platforms in a variety of environments, from forests to fusion reactors, for durations ranging from hours to weeks. The AutoInspect system combines robust mapping and localisation with graph-based autonomous navigation, mission execution, and scheduling, into a complete autonomous inspection and monitoring system. Code for the graph-based autonomy component will be made available at ori.ox.ac.uk/projects/autoinspect. In this article we describe in detail our most commonly-used configuration of AutoInspect: a Boston Dynamics Spot fitted with a custom compute and sensing payload, called Frontier. To highlight the flexibility of the AutoInspect system, we also describe its deployment with different hardware and software configurations. We evaluate AutoInspect’s performance in two long-term deployments on Spot at the Culham Centre for Fusion Energy in Oxfordshire, UK. The first deployment took place at a robotics test facility, spanning 49 days, including 14 uninterrupted days of autonomous operation. The second deployment covered 35 days in the torus hall of the Joint European Torus (JET) fusion reactor, and achieved 15 days of uninterrupted operation. This was the first ever deployment of a fully autonomous mobile robot in a fusion facility.


  • Task-Oriented Co-Design of Communication, Computing, and Control for Edge-Enabled Industrial Cyber-Physical Systems

    by Yufeng Diao, Yichi Zhang, Daniele De Martini, Philip Guodong Zhao, Emma Liying Li

    This paper proposes a task-oriented co-design framework that integrates communication, computing, and control to address the key challenges of bandwidth limitations, noise interference, and latency in mission-critical industrial Cyber-Physical Systems (CPS). To improve communication efficiency and robustness, we design a task-oriented Joint Source-Channel Coding (JSCC) using Information Bottleneck (IB) to enhance data transmission efficiency by prioritizing task-specific information. To mitigate the perceived End-to-End (E2E) delays, we develop a Delay-Aware Trajectory-Guided Control Prediction (DTCP) strategy that integrates trajectory planning with control prediction, predicting commands based on E2E delay. Moreover, the DTCP is co-designed with task-oriented JSCC, focusing on transmitting task-specific information for timely and reliable autonomous driving. Experimental results in the CARLA simulator demonstrate that, under an E2E delay of 1 second (20 time slots), the proposed framework achieves a driving score of 48.12, which is 31.59 points higher than using Better Portable Graphics (BPG) while reducing bandwidth usage by 99.19%.


  • The Oxford RobotCycle Project: A Multimodal Urban Cycling Dataset for Assessing the Safety of Vulnerable Road Users

    by Efimia Panagiotaki, Divya Thuremella, Jumana Baghabrah, Samuel Sze, Lanke Frank Tarimo Fu, Benjamin Hardin, Tyler Reinmund, Tobit Flatscher, Daniel Marques, Chris Prahacs, et al.

    The Oxford RobotCycle Project is a novel initiative aiming to understand how road and traffic infrastructure influence road users’ behaviour, affecting cyclists’ journeys and safety. By leveraging state-of-the-art technology and methods used in Autonomous Vehicles (AVs), this project introduces a novel multimodal dataset, capturing dynamic cycling data in complex and diverse urban traffic environments. The dataset consists of range, visual, and inertial sensors, mounted on a backpack, and eye-gaze tracking glasses, coupled with an analysis of road infrastructure and interactions with other road users. Enhanced by annotated maps, reconstructed 3D pointclouds, and a detailed ontology capturing static and dynamic agents and their relations, the dataset provides a comprehensive framework for analysing and understanding traffic dynamics. Heatmaps derived from the cyclists’ vision reveal attention patterns and focal points during various traffic scenarios. We also analyse traffic interactions and risk, either perceived or actual, and correlate it with road infrastructure and traffic volumes. To complement the dataset we also provide a complete set of tools for risk and traffic analysis, visualisation, automatic calibration, and data annotation. The dataset can also be used to evaluate the robustness of odometry estimation methods, due to the highly dynamic cyclist movements. Combining multimodal data with traffic and risk analysis, the Oxford RobotCycle Project facilitates identifying safety-critical scenarios to derive actionable insights for safer, cyclist-friendly road design. This work contributes towards improving cycling safety, enhancing urban mobility, and supporting sustainable transportation initiatives.


  • Open Access

    Task-Oriented Co-Design of Communication, Computing, and Control for Edge-Enabled Industrial Cyber-Physical Systems

    by Yufeng Diao, Yichi Zhang, Daniele De Martini, Philip Guodong Zhao, Emma Liying Li

    This paper proposes a task-oriented co-design framework that integrates communication, computing, and control to address the key challenges of bandwidth limitations, noise interference, and latency in mission-critical industrial Cyber-Physical Systems (CPS). To improve communication efficiency and robustness, we design a task-oriented Joint Source-Channel Coding (JSCC) using Information Bottleneck (IB) to enhance data transmission efficiency by prioritizing task-specific information. To mitigate the perceived End-to-End (E2E) delays, we develop a Delay-Aware Trajectory-Guided Control Prediction (DTCP) strategy that integrates trajectory planning with control prediction, predicting commands based on E2E delay. Moreover, the DTCP is co-designed with task-oriented JSCC, focusing on transmitting task-specific information for timely and reliable autonomous driving. Experimental results in the CARLA simulator demonstrate that, under an E2E delay of 1 second (20 time slots), the proposed framework achieves a driving score of 48.12, which is 31.59 points higher than using Better Portable Graphics (BPG) while reducing bandwidth usage by 99.19%.

  • OORD: The Oxford Offroad Radar Dataset

    by Matthew Gadd, Daniele De Martini, Oliver Bartlett, Paul Murcutt, Matt Towlson, Matthew Widojo, Valentina Muşat, Luke Robinson, Efimia Panagiotaki, Georgi Pramatarov, et al.

    There is a growing academic interest as well as commercial exploitation of millimetre-wave scanning radar for autonomous vehicle localisation and scene understanding. Although several datasets to support this research area have been released, they are primarily focused on urban or semi-urban environments. Nevertheless, rugged offroad deployments are important application areas which also present unique challenges and opportunities for this sensor technology. Therefore, the Oxford Offroad Radar Dataset (OORD) presents data collected in the rugged Scottish highlands in extreme weather. The radar data we offer to the community are accompanied by GPS/INS reference – to further stimulate research in radar place recognition. In total we release over 90 GiB of radar scans as well as GPS and IMU readings by driving a diverse set of four routes over $11$ forays, totalling approximately 154 km of rugged driving. This is an area increasingly explored in literature, and we therefore present and release examples of recent open-sourced radar place recognition systems and their performance on our dataset. This includes a learned neural network, the weights of which we also release. The data and tools are made freely available to the community at oxford-robotics-institute.github.io/oord-dataset