Paul H. J. Kelly
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Paul H. J. Kelly has a total of 84 co-authors. Here are the top co-authors by number of shared publications:
- Riku Murai
Imperial College London
7 shared publications - Andrew J. Davison
Imperial College London
6 shared publications - Gerard Gorman
Imperial College London London UK
6 shared publications - Georgios Rokos
Imperial College London
4 shared publications - Sajad Saeedi
University College London, London, U.K.
3 shared publications - Sajad Saeedi
Department of Mechanical and Industrial Engineering, Toronto Metropolitan University, Toronto, Canada
3 shared publications
Latest Publications
- 16 September 2026
- Published by Springer Nature
- p. 215-232
Triangle Splatting SLAM
- 15 September 2026
- Published by Springer Nature
- p. 58-75
MLP Splatting: Object-Centric Neural Fields
- 12 September 2026
- Published by Springer Nature
- p. 591-608
PixVOD: Pixel-Distributed Direct Visual Odometry and Depth Estimation
- 12 September 2026
- Published by Springer Nature
- p. 591-608
PixVOD: Pixel-Distributed Direct Visual Odometry and Depth Estimation
- Open Access
Versatile Cross-platform Compilation Toolchain for Schrödinger-style Quantum Circuit Simulation
While existing quantum hardware resources have limited availability and reliability, there is a growing demand for exploring and verifying quantum algorithms. Efficient classical simulators for high-performance quantum simulation are critical to meeting this demand. However, due to the vastly varied characteristics of classical hardware, implementing hardware-specific optimizations for different hardware platforms is challenging. To address such needs, we propose CAST (Cross-platform Adaptive Schrödiner-style Simulation Toolchain), a novel compilation toolchain with cross-platform (CPU and Nvidia GPU) optimization and high-performance backend supports. CAST exploits a novel sparsity-aware gate fusion algorithm that automatically selects the best fusion strategy and backend configuration for targeted hardware platforms. CAST also aims to offer versatile and high-performance backend for different hardware platforms. To this end, CAST provides an LLVM IR-based vectorization optimization for various CPU architectures and instruction sets, as well as a PTX-based code generator for Nvidia GPU support. We benchmark CAST against IBM Qiskit, Google QSimCirq, Nvidia cuQuantum backend, and other high-performance simulators. On various 32-qubit CPU-based benchmarks, CAST is able to achieve up to 8.03x speedup than Qiskit. On various 30-qubit GPU-based benchmarks, CAST is able to achieve up to 39.3x speedup than Nvidia cuQuantum backend.
Open AccessVersatile Cross-platform Compilation Toolchain for Schrödinger-style Quantum Circuit Simulation
While existing quantum hardware resources have limited availability and reliability, there is a growing demand for exploring and verifying quantum algorithms. Efficient classical simulators for high-performance quantum simulation are critical to meeting this demand. However, due to the vastly varied characteristics of classical hardware, implementing hardware-specific optimizations for different hardware platforms is challenging. To address such needs, we propose CAST (Cross-platform Adaptive Schrödiner-style Simulation Toolchain), a novel compilation toolchain with cross-platform (CPU and Nvidia GPU) optimization and high-performance backend supports. CAST exploits a novel sparsity-aware gate fusion algorithm that automatically selects the best fusion strategy and backend configuration for targeted hardware platforms. CAST also aims to offer versatile and high-performance backend for different hardware platforms. To this end, CAST provides an LLVM IR-based vectorization optimization for various CPU architectures and instruction sets, as well as a PTX-based code generator for Nvidia GPU support. We benchmark CAST against IBM Qiskit, Google QSimCirq, Nvidia cuQuantum backend, and other high-performance simulators. On various 32-qubit CPU-based benchmarks, CAST is able to achieve up to 8.03x speedup than Qiskit. On various 30-qubit GPU-based benchmarks, CAST is able to achieve up to 39.3x speedup than Nvidia cuQuantum backend.
- Open Access
- 27 April 2024
- Published by Association for Computing Machinery (ACM)
- p. 38-56
A shared compilation stack for distributed-memory parallelism in stencil DSLs
- 27 April 2024
- Published by Association for Computing Machinery (ACM)
- p. 38-56
Open AccessA shared compilation stack for distributed-memory parallelism in stencil DSLs
- Open Access
Gaussian Splatting SLAM
We present the first application of 3D Gaussian Splatting in monocular SLAM, the most fundamental but the hardest setup for Visual SLAM. Our method, which runs live at 3fps, utilises Gaussians as the only 3D representation, unifying the required representation for accurate, efficient tracking, mapping, and high-quality rendering. Designed for challenging monocular settings, our approach is seamlessly extendable to RGB-D SLAM when an external depth sensor is available. Several innovations are required to continuously reconstruct 3D scenes with high fidelity from a live camera. First, to move beyond the original 3DGS algorithm, which requires accurate poses from an offline Structure from Motion (SfM) system, we formulate camera tracking for 3DGS using direct optimisation against the 3D Gaussians, and show that this enables fast and robust tracking with a wide basin of convergence. Second, by utilising the explicit nature of the Gaussians, we introduce geometric verification and regularisation to handle the ambiguities occurring in incremental 3D dense reconstruction. Finally, we introduce a full SLAM system which not only achieves state-of-the-art results in novel view synthesis and trajectory estimation but also reconstruction of tiny and even transparent objects.
Open AccessGaussian Splatting SLAM
We present the first application of 3D Gaussian Splatting in monocular SLAM, the most fundamental but the hardest setup for Visual SLAM. Our method, which runs live at 3fps, utilises Gaussians as the only 3D representation, unifying the required representation for accurate, efficient tracking, mapping, and high-quality rendering. Designed for challenging monocular settings, our approach is seamlessly extendable to RGB-D SLAM when an external depth sensor is available. Several innovations are required to continuously reconstruct 3D scenes with high fidelity from a live camera. First, to move beyond the original 3DGS algorithm, which requires accurate poses from an offline Structure from Motion (SfM) system, we formulate camera tracking for 3DGS using direct optimisation against the 3D Gaussians, and show that this enables fast and robust tracking with a wide basin of convergence. Second, by utilising the explicit nature of the Gaussians, we introduce geometric verification and regularisation to handle the ambiguities occurring in incremental 3D dense reconstruction. Finally, we introduce a full SLAM system which not only achieves state-of-the-art results in novel view synthesis and trajectory estimation but also reconstruction of tiny and even transparent objects.
- 10 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 9 (3) , 2136-2143
Distributed Simultaneous Localisation and Auto-Calibration Using Gaussian Belief Propagation
We present a novel scalable, fully distributed, and online method for simultaneous localisation and extrinsic calibration for multi-robot setups. Individual a priori unknown robot poses are probabilistically inferred as robots sense each other while simultaneously calibrating their sensors and markers extrinsic using Gaussian Belief Propagation. In the presented experiments, we show how our method not only yields accurate robot localisation and auto-calibration but also is able to perform under challenging circumstances such as highly noisy measurements, significant communication failures or limited communication range.
- 10 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 9 (3) , 2136-2143
Distributed Simultaneous Localisation and Auto-Calibration Using Gaussian Belief Propagation
We present a novel scalable, fully distributed, and online method for simultaneous localisation and extrinsic calibration for multi-robot setups. Individual a priori unknown robot poses are probabilistically inferred as robots sense each other while simultaneously calibrating their sensors and markers extrinsic using Gaussian Belief Propagation. In the presented experiments, we show how our method not only yields accurate robot localisation and auto-calibration but also is able to perform under challenging circumstances such as highly noisy measurements, significant communication failures or limited communication range.
- Top Cited Papers
- 16 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Robotics
- Vol. 40 (15523098) , 121-138
A Robot Web for Distributed Many-Device Localization
We show that a distributed network of robots or other devices which make measurements of each other can collaborate to globally localize via efficient ad hoc peer-to-peer communication. Our Robot Web solution is based on Gaussian belief propagation (GBP) on the fundamental nonlinear factor graph describing the probabilistic structure of all of the observations robots make internally or of each other, and is flexible for any type of robot, motion or sensor. We define a simple and efficient communication protocol which can be implemented by the publishing and reading of web pages or other asynchronous communication technologies. We show in simulations with up to 1000 robots interacting in arbitrary patterns that our solution convergently achieves global accuracy as accurate as a centralized nonlinear factor graph solver while operating with high distributed efficiency of computation and communication. Via the use of robust factors in GBP, our method is tolerant to a high percentage of faulty sensor measurements or dropped communication packets. Furthermore, we showcase that the system operates on real robots with limited onboard computational resources.
- 16 October 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Robotics
- Vol. 40 (15523098) , 121-138
Top Cited PapersA Robot Web for Distributed Many-Device Localization
We show that a distributed network of robots or other devices which make measurements of each other can collaborate to globally localize via efficient ad hoc peer-to-peer communication. Our Robot Web solution is based on Gaussian belief propagation (GBP) on the fundamental nonlinear factor graph describing the probabilistic structure of all of the observations robots make internally or of each other, and is flexible for any type of robot, motion or sensor. We define a simple and efficient communication protocol which can be implemented by the publishing and reading of web pages or other asynchronous communication technologies. We show in simulations with up to 1000 robots interacting in arbitrary patterns that our solution convergently achieves global accuracy as accurate as a centralized nonlinear factor graph solver while operating with high distributed efficiency of computation and communication. Via the use of robust factors in GBP, our method is tolerant to a high percentage of faulty sensor measurements or dropped communication packets. Furthermore, we showcase that the system operates on real robots with limited onboard computational resources.
- Open Access
- 22 July 2023
- Published by Springer Nature in Autonomous Robots
- Vol. 47 (8) , 1579-1592
High-frame rate homography and visual odometry by tracking binary features from the focal plane
Robotics faces a long-standing obstacle in which the speed of the vision system’s scene understanding is insufficient, impeding the robot’s ability to perform agile tasks. Consequently, robots must often rely on interpolation and extrapolation of the vision data to accomplish tasks in a timely and effective manner. One of the primary reasons for these delays is the analog-to-digital conversion that occurs on a per-pixel basis across the image sensor, along with the transfer of pixel-intensity information to the host device. This results in significant delays and power consumption in modern visual processing pipelines. The SCAMP-5—a general-purpose Focal-plane Sensor-processor array (FPSP)—used in this research performs computations in the analog domain prior to analog-to-digital conversion. By extracting features from the image on the focal plane, the amount of data that needs to be digitised and transferred is reduced. This allows for a high frame rate and low energy consumption for the SCAMP-5. The focus of our work is on localising the camera within the scene, which is crucial for scene understanding and for any downstream robotics tasks. We present a localisation system that utilise the FPSP in two parts. First, a 6-DoF odometry system is introduced, which efficiently estimates its position against a known marker at over 400 FPS. Second, our work is extended to implement BIT-VO—6-DoF visual odometry system which operates under an unknown natural environment at 300 FPS.
- 22 July 2023
- Published by Springer Nature in Autonomous Robots
- Vol. 47 (8) , 1579-1592
Open AccessHigh-frame rate homography and visual odometry by tracking binary features from the focal plane
Robotics faces a long-standing obstacle in which the speed of the vision system’s scene understanding is insufficient, impeding the robot’s ability to perform agile tasks. Consequently, robots must often rely on interpolation and extrapolation of the vision data to accomplish tasks in a timely and effective manner. One of the primary reasons for these delays is the analog-to-digital conversion that occurs on a per-pixel basis across the image sensor, along with the transfer of pixel-intensity information to the host device. This results in significant delays and power consumption in modern visual processing pipelines. The SCAMP-5—a general-purpose Focal-plane Sensor-processor array (FPSP)—used in this research performs computations in the analog domain prior to analog-to-digital conversion. By extracting features from the image on the focal plane, the amount of data that needs to be digitised and transferred is reduced. This allows for a high frame rate and low energy consumption for the SCAMP-5. The focus of our work is on localising the camera within the scene, which is crucial for scene understanding and for any downstream robotics tasks. We present a localisation system that utilise the FPSP in two parts. First, a 6-DoF odometry system is introduced, which efficiently estimates its position against a known marker at over 400 FPS. Second, our work is extended to implement BIT-VO—6-DoF visual odometry system which operates under an unknown natural environment at 300 FPS.
- Open Access
- 14 March 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Parallel and Distributed Systems
- Vol. 34 (5) , 1594-1608
Precise Event Sampling on AMD Versus Intel: Quantitative and Qualitative Comparison
Precise event sampling is a profiling feature in commodity processors that can sample hardware events and accurately locate the instructions that trigger the events. This feature has been used in a large number of tools to detect application performance issues. Although precise event sampling is readily supported in modern multicore architectures, vendor supports exhibit great differences that affect their accuracy, stability, overhead, and functionality. This work presents the most comprehensive study to date on benchmarking the event sampling features of Intel PEBS and AMD IBS and performs in-depth analysis on key differences through series of microbenchmarks. Our qualitative and quantitative analysis shows that PEBS allows finer-grained and more accurate sampling of hardware events, while IBS offers richer set of information at each sample though it suffers from lower accuracy and stability. Moreover, OS signal delivery, which is a common method used by the profiling software, introduces significant time overhead to the original overhead incurred by the hardware mechanisms in both PEBS and IBS. We also found that both PEBS and IBS have bias in sampling events across multiple different locations in a code. Lastly, we demonstrate how our findings on microbenchmarks under different thread counts hold for a full-fledged profiling tool that runs on the state-of-the-art Intel and AMD machines. Overall our detailed comparisons serve as a great reference and provide invaluable information for hardware designers and profiling tool developers.
- 14 March 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Parallel and Distributed Systems
- Vol. 34 (5) , 1594-1608
Open AccessPrecise Event Sampling on AMD Versus Intel: Quantitative and Qualitative Comparison
Precise event sampling is a profiling feature in commodity processors that can sample hardware events and accurately locate the instructions that trigger the events. This feature has been used in a large number of tools to detect application performance issues. Although precise event sampling is readily supported in modern multicore architectures, vendor supports exhibit great differences that affect their accuracy, stability, overhead, and functionality. This work presents the most comprehensive study to date on benchmarking the event sampling features of Intel PEBS and AMD IBS and performs in-depth analysis on key differences through series of microbenchmarks. Our qualitative and quantitative analysis shows that PEBS allows finer-grained and more accurate sampling of hardware events, while IBS offers richer set of information at each sample though it suffers from lower accuracy and stability. Moreover, OS signal delivery, which is a common method used by the profiling software, introduces significant time overhead to the original overhead incurred by the hardware mechanisms in both PEBS and IBS. We also found that both PEBS and IBS have bias in sampling events across multiple different locations in a code. Lastly, we demonstrate how our findings on microbenchmarks under different thread counts hold for a full-fledged profiling tool that runs on the state-of-the-art Intel and AMD machines. Overall our detailed comparisons serve as a great reference and provide invaluable information for hardware designers and profiling tool developers.