World modelling is essential for understanding and predicting the dynamics of complex systems by learning both spatial and temporal dependencies. However, current frameworks, such as Transformers and selective state-space models like Mambas, exhibit limitations in efficiently encoding spatial and temporal structures, particularly in scenarios requiring long-term high-dimensional sequence modelling. To address these issues, we propose a novel recurrent framework, the FACTored State-space (FACTS) model, for spatial-temporal world modelling. The FACTS framework constructs a graph-structured memory with a routing mechanism that learns permutable memory representations, ensuring invariance to input permutations while adapting through selective state-space propagation. Furthermore, FACTS supports parallel computation of high-dimensional sequences. We empirically evaluate FACTS across diverse tasks, including multivariate time series forecasting, object-centric world modelling, and spatial-temporal graph prediction, demonstrating that it consistently outperforms or matches specialised state-of-the-art models, despite its general-purpose world modelling design.

Videos capture the evolution of continuous dynamical systems over time in the form of discrete image sequences. Recently, video generation models have been widely used in robotic research. However, generating controllable videos from image-text pairs is an important yet underexplored research topic in both robotic and computer vision communities. This paper introduces an innovative and elegant framework named TiV-ODE, formulating this task as modeling the dynamical system in a continuous space. Specifically, our framework leverages the ability of Neural Ordinary Differential Equations (Neural ODEs) to model the complex dynamical system depicted by videos as a nonlinear ordinary differential equation. The resulting framework offers control over the generated videos’ dynamics, content, and frame rate, a feature not provided by previous methods. Experiments demonstrate the ability of the proposed method to generate highly controllable and visually consistent videos and its capability of modeling dynamical systems. Overall, this work is a significant step towards developing advanced controllable video generation models that can handle complex and dynamic scenes.

Current datasets for computer vision-based action recognition and localization cover a wide range of classes and challenging scenarios. However, these datasets don't cater to healthcare applications that involve long-term monitoring, tracking minor changes in movements over time for healthcare purposes, or completely modeling a specific human behavior that includes multiple sub-actions. Specifically, there are no existing datasets for research on either health monitoring on atomic-action-based eating behavior or for a full range of eating sub-actions that fully segment the main action. Addressing these gaps is valuable for extending research on the health monitoring of elderly people and is needed for creating richer and more complete descriptions of actions. This paper introduces a new benchmark dataset named EatSense that targets both the computer vision and healthcare communities and fills in the aforementioned gaps. EatSense is recorded while a person eats in an uncontrolled dining setting. The key features of EatSense are the introduction of challenging atomic actions for action recognition, the significantly diverse durations of actions that make it difficult for current temporal action localization frameworks to localize, the capability to model comprehensive eating behavior in terms of a sequence of action-based behaviors, and the simulation of minor variations in motion or performance. We conduct extensive experiments on EatSense with baseline deep learning-based approaches for benchmarking and hand-crafted feature-based approaches for explainable applications. We believe this dataset will benefit future researchers in building robust temporal action localization networks, behavior recognition, and performance assessment models for eating.

Learning object-centric scene representations is essential for attaining structural understanding and abstraction of complex scenes. Yet, as current approaches for unsupervised object-centric representation learning are built upon either a stationary observer assumption or a static scene assumption, they often: i) suffer single-view spatial ambiguities, or ii) infer incorrectly or inaccurately object representations from dynamic scenes. To address this, we propose Dynamics-aware Multi-Object Network (DyMON), a method that broadens the scope of multi-view object-centric representation learning to dynamic scenes. We train DyMON on multi-view-dynamic-scene data and show that DyMON learns---without supervision---to factorize the entangled effects of observer motions and scene object dynamics from a sequence of observations, and constructs scene object spatial representations suitable for rendering at arbitrary times (querying across time) and from arbitrary viewpoints (querying across space). We also show that the factorized scene representations (w.r.t. objects) support querying about a single object by space and time independently.

Learning object-centric representations of multi-object scenes is a promising approach towards machine intelligence, facilitating high-level reasoning and control from visual sensory data. However, current approaches for \textit{unsupervised object-centric scene representation} are incapable of aggregating information from multiple observations of a scene. As a result, these ``single-view'' methods form their representations of a 3D scene based only on a single 2D observation (view). Naturally, this leads to several inaccuracies, with these methods falling victim to single-view spatial ambiguities. To address this, we propose \textit{The Multi-View and Multi-Object Network (MulMON)}---a method for learning accurate, object-centric representations of multi-object scenes by leveraging multiple views. In order to sidestep the main technical difficulty of the \textit{multi-object-multi-view} scenario---maintaining object correspondences across views---MulMON iteratively updates the latent object representations for a scene over multiple views. To ensure that these iterative updates do indeed aggregate spatial information to form a complete 3D scene understanding, MulMON is asked to predict the appearance of the scene from novel viewpoints during training. Through experiments we show that MulMON better-resolves spatial ambiguities than single-view methods---learning more accurate and disentangled object representations---and also achieves new functionality in predicting object segmentations for novel viewpoints.

Accurately registering point clouds from a cheap low resolution sensor is a challenging task. Existing rigid registration methods failed to use the physical 3D uncertainty distribution of each point from a real sensor in the dynamic alignment process. It is mainly because the uncertainty model for a point is static and invariant and it is hard to describe the change of these physical uncertainty models in different views. Additionally, the existing Gaussian mixture alignment architecture cannot efficiently implement these dynamic changes. This paper proposes a simple architecture combining error estimation from sample covariances and dual dynamic global probability alignment using the convolution of uncertainty-based Gaussian Mixture Models (GMM) from point clouds. Firstly, we propose an efficient way to describe the change of each 3D uncertainty model, which represents the structure of the point cloud much better. Unlike the invariant GMM (representing a fixed point cloud) in traditional Gaussian mixture alignment, we use two uncertainty-based GMMs that change and interact with each other in each iteration. In order to have a wider basin of convergence than other local algorithms, we design a more robust energy function by convolving efficiently the two GMMs over the whole 3D space. Tens of thousands of trials have been conducted on hundreds of models from multiple datasets to demonstrate the proposed method’s superior performance compared with the current state-of-the-art methods. All the materials including our code is available from https://github. com/Canpu999/DUGMA.

Visual servoing is a well-known task in robotics. However, there are still challenges when multiple visual sources are combined to accurately guide the robot or occlusions appear. In this paper we present a novel visual servoing approach using hybrid multi-camera input data to lead a robot arm accurately to dynamically moving target points in the presence of partial occlusions. The approach uses four RGBD sensors as Eye-to-Hand (EtoH) visual input, and an arm-mounted stereo camera as Eye-in-Hand (EinH). A Master supervisor task selects between using the EtoH or the EinH, depending on the distance between the robot and target. The Master also selects the subset of EtoH cameras that best perceive the target. When the EinH sensor is used, if the target becomes occluded or goes out of the sensor's view-frustum, the Master switches back to the EtoH sensors to re-track the object. Using this adaptive visual input data, the robot is then controlled using an iterative planner that uses position, orientation and joint configuration to estimate the trajectory. Since the target is dynamic, this trajectory is updated every time-step. Experiments show good performance in four different situations: tracking a ball, targeting a bulls-eye, guiding a straw to a mouth and delivering an item to a moving hand. The experiments cover both simple situations such as a ball that is mostly visible from all cameras, and more complex situations such as the mouth which is partially occluded from some of the sensors.