• Novel end-to-end causal decision-making framework for autonomous robots. • Integration of causal discovery and reasoning within the ROS ecosystem. • PeopleFlow: a context-sensitive Gazebo simulator for human-robot interaction. • Causal routing significantly improves robot safety and battery efficiency. • Preemptive task refusal based on causal battery consumption estimates. The growing integration of robots in shared environments—such as warehouses, shopping centres, and hospitals—demands a deep understanding of the underlying dynamics and human behaviours, including how, when, and where individuals engage in various activities and interactions. This knowledge goes beyond simple correlation studies and requires a more comprehensive causal analysis. By leveraging causal inference to model cause-and-effect relationships, we can better anticipate critical environmental factors and enable autonomous robots to plan and execute tasks more effectively. To this end, we propose a novel causality-based decision-making framework that reasons over a learned causal model to assist the robot in deciding when and how to complete a given task. In the examined use case—i.e., a warehouse shared with people—we exploit the causal model to estimate battery usage and human obstructions as factors influencing the robot’s task execution. This reasoning framework supports the robot in making informed decisions about task timing and strategy. To achieve this, we developed also PeopleFlow, a new Gazebo-based simulator designed to model context-sensitive human-robot spatial interactions in shared workspaces. PeopleFlow features realistic human and robot trajectories influenced by contextual factors such as time, environment layout, and robot state, and can simulate a large number of agents. While the simulator is general-purpose, in this paper we focus on a warehouse-like environment as a case study, where we conduct an extensive evaluation benchmarking our causal approach against a non-causal baseline. Our findings demonstrate the efficacy of the proposed solutions, highlighting how causal reasoning enables autonomous robots to operate more efficiently and safely in dynamic environments shared with humans.

Semi-supervised teacher-student pseudo-labelling improves instance segmentation by exploiting unlabelled data, where a teacher network, trained with a small annotated dataset, generates pseudo labels for the remaining data, to train the student model. However, mask selection typically relies heavily on the class confidence scores. In single-class settings these scores saturate, offering little discrimination between masks. In this work we propose a mask symmetry score that evaluates logits from the mask prediction head, enabling more reliable pseudo-label selection without architectural changes. Evaluations on both CNN- and Transformer-based models show our method outperforms state-of-the-art approaches on a real-world agri-robotic dataset of densely clustered potato tubers.

Deploying robots in human-shared environments requires a deep understanding of how nearby agents and objects interact. Employing causal inference to model cause-and-effect relationships facilitates the prediction of human behaviours and enables the anticipation of robot interventions. However, a significant challenge arises due to the absence of implementation of existing causal discovery methods within the ROS ecosystem, the standard de-facto framework in robotics, hindering effective utilisation on real robots. To bridge this gap, in our previous work we proposed ROS-Causal, a ROS-based framework designed for onboard data collection and causal discovery in human-robot spatial interactions. In this work, we present an experimental evaluation of ROS-Causal both in simulation and on a new dataset of human-robot spatial interactions in a lab scenario, to assess its performance and effectiveness. Our analysis demonstrates the efficacy of this approach, showcasing how causal models can be extracted directly onboard by robots during data collection. The online causal models generated from the simulation are consistent with those from lab experiments. These findings can help researchers to enhance the performance of robotic systems in shared environments, firstly by studying the causal relations between variables in simulation without real people, and then facilitating the actual robot deployment in real human environments. ROS-Causal: https://lcastri.github.io/roscausal

Reasoning on the context of human beings is crucial for many real-world applications especially for those deploying autonomous systems (e.g. robots). In this paper, we present a new approach for context reasoning to further advance the field of human motion prediction. We therefore propose a neuro-symbolic approach for human motion prediction (NeuroSyM), which weights differently the interactions in the neighbourhood by leveraging an intuitive technique for spatial representation called Qualitative Trajectory Calculus (QTC). The proposed approach is experimentally tested on medium and long term time horizons using two architectures from the state of art, one of which is a baseline for human motion prediction and the other is a baseline for generic multivariate time-series prediction. Six datasets of challenging crowded scenarios, collected from both fixed and mobile cameras, were used for testing. Experimental results show that the NeuroSyM approach outperforms in most cases the baseline architectures in terms of prediction accuracy.

Deploying service robots in our daily life, whether in restaurants, warehouses or hospitals, calls for the need to reason on the interactions happening in dense and dynamic scenes. In this paper, we present and benchmark three new approaches to model and predict multi-agent interactions in dense scenes, including the use of an intuitive qualitative representation. The proposed solutions take into account static and dynamic context to predict individual interactions. They exploit an input- and a temporal-attention mechanism, and are tested on medium and long-term time horizons. The first two approaches integrate different relations from the so-called Qualitative Trajectory Calculus (QTC) within a stateof-the-art deep neural network to create a symbol-driven neural architecture for predicting spatial interactions. The third approach implements a purely data-driven network for motion prediction, the output of which is post-processed to predict QTC spatial interactions. Experimental results on a popular robot dataset of challenging crowded scenarios show that the purely data-driven prediction approach generally outperforms the other two. The three approaches were further evaluated on a different but related human scenarios to assess their generalisation capability.

Collision detection is one of the most challenging tasks for unmanned aerial vehicles (UAVs). This is especially true for small or micro-UAVs due to their limited computational power. In nature, flying insects with compact and simple visual systems demonstrate their remarkable ability to navigate and avoid collision in complex environments. A good example of this is provided by locusts. They can avoid collisions in a dense swarm through the activity of a motion-based visual neuron called the Lobula giant movement detector (LGMD). The defining feature of the LGMD neuron is its preference for looming. As a flying insect's visual neuron, LGMD is considered to be an ideal basis for building UAV's collision detecting system. However, existing LGMD models cannot distinguish looming clearly from other visual cues, such as complex background movements caused by UAV agile flights. To address this issue, we proposed a new model implementing distributed spatial-temporal synaptic interactions, which is inspired by recent findings in locusts' synaptic morphology. We first introduced the locally distributed excitation to enhance the excitation caused by visual motion with preferred velocities. Then, radially extending temporal latency for inhibition is incorporated to compete with the distributed excitation and selectively suppress the nonpreferred visual motions. This spatial-temporal competition between excitation and inhibition in our model is, therefore, tuned to preferred image angular velocity representing looming rather than background movements with these distributed synaptic interactions. Systematic experiments have been conducted to verify the performance of the proposed model for UAV agile flights. The results have demonstrated that this new model enhances the looming selectivity in complex flying scenes considerably and has the potential to be implemented on embedded collision detection systems for small or micro-UAVs.

Robots learning a new manipulation task from a small amount of demonstrations are increasingly demanded in different workspaces. A classifier model assessing the quality of actions can predict the successful completion of a task, which can be used by intelligent agents for action-selection. This paper presents a novel classifier that learns to classify task completion only from a few demonstrations. We carry out a comprehensive comparison of different neural classifiers, e.g. fully connected-based, fully convolutional-based, sequence2sequence-based, and domain adaptation-based classification. We also present a new dataset including five robot manipulation tasks, which is publicly available. We compared the performances of our novel classifier and the existing models using our dataset and the MIME dataset. The results suggest domain adaptation and timing-based features improve success prediction. Our novel model, i.e. fully convolutional neural network with domain adaptation and timing features, achieves an average classification accuracy of 97.3% and 95.5% across tasks in both datasets whereas state-of-the-art classifiers without domain adaptation and timing-features only achieve 82.4% and 90.3%, respectively.

Autonomous vehicles (AVs) must share space with pedestrians, both in carriageway cases such as cars at pedestrian crossings and off-carriageway cases such as delivery vehicles navigating through crowds on pedestrianized high-streets. Unlike static obstacles, pedestrians are active agents with complex, interactive motions. Planning AV actions in the presence of pedestrians thus requires modelling of their probable future behavior as well as detecting and tracking them. This narrative review article is Part I of a pair, together surveying the current technology stack involved in this process, organising recent research into a hierarchical taxonomy ranging from low-level image detection to high-level psychology models, from the perspective of an AV designer. This self-contained Part I covers the lower levels of this stack, from sensing, through detection and recognition, up to tracking of pedestrians. Technologies at these levels are found to be mature and available as foundations for use in high-level systems, such as behavior modelling, prediction and interaction control.

Autonomous vehicles (AVs) must share space with pedestrians, both in carriageway cases such as cars at pedestrian crossings and off-carriageway cases such as delivery vehicles navigating through crowds on pedestrianized high-streets. Unlike static obstacles, pedestrians are active agents with complex, interactive motions. Planning AV actions in the presence of pedestrians thus requires modelling of their probable future behavior as well as detecting and tracking them. This narrative review article is Part II of a pair, together surveying the current technology stack involved in this process, organising recent research into a hierarchical taxonomy ranging from low-level image detection to high-level psychological models, from the perspective of an AV designer. This self-contained Part II covers the higher levels of this stack, consisting of models of pedestrian behavior, from prediction of individual pedestrians' likely destinations and paths, to game-theoretic models of interactions between pedestrians and autonomous vehicles. This survey clearly shows that, although there are good models for optimal walking behavior, high-level psychological and social modelling of pedestrian behavior still remains an open research question that requires many conceptual issues to be clarified. Early work has been done on descriptive and qualitative models of behavior, but much work is still needed to translate them into quantitative algorithms for practical AV control.

Object detection plays a crucial role in the development of Electronic Travel Aids (ETAs), capable to guide a person with visual impairments towards a target object in an unknown indoor environment. In such a scenario, the object detector runs on a mobile device (e.g. smartphone) and needs to be fast, accurate, and, most importantly, lightweight. Nowadays, Deep Neural Networks (DNN) have become the state-of-the-art solution for object detection tasks, with many works improving speed and accuracy by proposing new architectures or extending existing ones. A common strategy is to use deeper networks to get higher performance, but that leads to a higher computational cost which makes it impractical to integrate them on mobile devices with limited computational power. In this work we compare different object detectors to find a suitable candidate to be implemented on ETAs, focusing on lightweight models capable of working in real-time on mobile devices with a good accuracy. In particular, we select two models: SSD Lite with Mobilenet V2 and Tiny-DSOD. Both models have been tested on the popular OpenImage dataset and a new dataset, named L-CAS Office dataset, collected to further test models' performance and robustness in a real scenario inspired by the actual perception challenges of a user with visual impairments.

With the increasing popularity of social media and smart devices, the face as one of the key biometrics becomes vital for person identification. Among those face recognition algorithms, video-based face recognition methods could make use of both temporal and spatial information just as humans do to achieve better classification performance. However, they cannot identify individuals when certain key facial areas, such as eyes or nose, are disguised by heavy makeup or rubber/digital masks. To this end, we propose a novel deep spiking neural network architecture in this paper. It takes dynamic facial movements, the facial muscle changes induced by speaking or other activities, as the sole input. An event-driven continuous spike-timing-dependent plasticity learning rule with adaptive thresholding is applied to train the synaptic weights. The experiments on our proposed video-based disguise face database (MakeFace DB) demonstrate that the proposed learning method performs very well, i.e., it achieves from 95% to 100% correct classification rates under various realistic experimental scenarios.

Human detection and tracking is an essential task for service robots, where the combined use of multiple sensors has potential advantages that are yet to be fully exploited. In this paper, we introduce a framework allowing a robot to learn a new 3D LiDAR-based human classifier from other sensors over time, taking advantage of a multisensor tracking system. The main innovation is the use of different detectors for existing sensors (i.e. RGB-D camera, 2D LiDAR) to train, online, a new 3D LiDAR-based human classifier based on a new “trajectory probability”. Our framework uses this probability to check whether new detection belongs to a human trajectory, estimated by different sensors and/or detectors, and to learn a human classifier in a semi-supervised fashion. The framework has been implemented and tested on a real-world dataset collected by a mobile robot. We present experiments illustrating that our system is able to effectively learn from different sensors and from the environment, and that the performance of the 3D LiDAR-based human classification improves with the number of sensors/detectors used.

Automated recognition of Activities of Daily Living allows to identify possible health problems and apply corrective strategies in Ambient Assisted Living (AAL). Activities of Daily Living analysis can provide very useful information for elder care and long-term care services. This paper presents an automated RGB-D video analysis system that recognises human ADLs activities, related to classical daily actions. The main goal is to predict the probability of an analysed subject action. Thus, abnormal behaviour can be detected. The activity detection and recognition is performed using an affordable RGB-D camera. Human activities, despite their unstructured nature, tend to have a natural hierarchical structure; for instance, generally making a coffee involves a three-step process of turning on the coffee machine, putting sugar in cup and opening the fridge for milk. Action sequence recognition is then handled using a discriminative Hidden Markov Model (HMM). RADiaL, a dataset with RGB-D images and 3D position of each person for training as well as evaluating the HMM, has been built and made publicly available.

Human detection and tracking are essential aspects to be considered in service robotics, as the robot often shares its workspace and interacts closely with humans. This paper presents an online learning framework for human classification in 3D LiDAR scans, taking advantage of robust multi-target tracking to avoid the need for data annotation by a human expert. The system learns iteratively by retraining a classifier online with the samples collected by the robot over time. A novel aspect of our approach is that errors in training data can be corrected using the information provided by the 3D LiDAR-based tracking. In order to do this, an efficient 3D cluster detector of potential human targets has been implemented. We evaluate the framework using a new 3D LiDAR dataset of people moving in a large indoor public space, which is made available to the research community. The experiments analyse the real-time performance of the cluster detector and show that our online learned human classifier matches and in some cases outperforms its offline version.

We present a system for temporal detection of social interactions. Many of the works until now have succeeded in recognising activities from clipped videos in datasets, but for robotic applications, it is important to be able to move to more realistic data. For this reason, the proposed approach temporally detects intervals where individual or social activity is occurring. Recognition of human activities is a key feature for analysing the human behaviour. In particular, recognition of social activities is useful to trigger human-robot interactions or to detect situations of potential danger. Based on that, this research has three goals: (1) define a new set of descriptors, which are able to characterise human interactions; (2) develop a computational model to segment temporal intervals with social interaction or individual behaviour; (3) provide a public dataset with RGB-D data with continuous stream of individual activities and social interactions. Results show that the proposed approach attained relevant performance with temporal segmentation of social activities.

Stress remains a significant social problem for individuals in modern societies. This paper presents a machine learning approach for the automatic detection of stress of people in a social situation by combining two sensor systems that capture physiological and social responses. We compare the performance using different classifiers including support vector machine, AdaBoost, and [Formula: see text]-nearest neighbor. Our experimental results show that by combining the measurements from both sensor systems, we could accurately discriminate between stressful and neutral situations during a controlled Trier social stress test (TSST). Moreover, this paper assesses the discriminative ability of each sensor modality individually and considers their suitability for real-time stress detection. Finally, we present an study of the most discriminative features for stress detection.

We investigate how incremental learning of long-term human activity patterns improves the accuracy of activity classification over time. Rather than trying to improve the classification methods themselves, we assume that they can take into account prior probabilities of activities occurring at a particular time. We use the classification results to build temporal models that can provide these priors to the classifiers. As our system gradually learns about typical patterns of human activities, the accuracy of activity classification improves, which results in even more accurate priors. Two datasets collected over several months containing hand-annotated activity in residential and office environments were chosen to evaluate the approach. Several types of temporal models were evaluated for each of these datasets. The results indicate that incremental learning of daily routines leads to a significant improvement in activity classification.

We address possible solutions for a practical application of Markov Logic Networks to online activity recognition, based on domotic sensors, to be used for monitoring elderly with mild cognitive impairments. Our system has to provide responsive information about user activities throughout the day, so different inference engines are tested. We use an abstraction layer to gather information from commercial domotic sensors. Sensor events are stored using a non-relational database. Using this database, evidences are built to query a logic network about current activities. Markov Logic Networks are able to deal with uncertainty while keeping a structured knowledge. This makes them a suitable tool for ambient sensors based inference. However, in their previous application, inferences are usually made offline. Time is a relevant constrain in our system and hence logic networks are designed here accordingly. We compare in this work different engines to model a Markov Logic Network suitable for such circumstances. Results show some insights about how to design a low latency logic network and which kind of solutions should be avoided.

Social activity based on body motion is a key feature for non-verbal and physical behavior defined as function for communicative signal and social interaction between individuals. Social activity recognition is important to study human-human communication and also human-robot interaction. Based on that, this research has threefold goals: (1) recognition of social behavior (e.g. human-human interaction) using a probabilistic approach that merges spatio-temporal features from individual bodies and social features from the relationship between two individuals; (2) learn priors based on physical proximity between individuals during an interaction using proxemics theory to feed a probabilistic ensemble of activity classifiers; and (3) provide a public dataset with RGB-D data of social daily activities including risk situations useful to test approaches for assisted living, since this type of dataset is still missing. Results show that using the proposed approach designed to merge features with different semantics and proximity priors improves the classification performance in terms of precision, recall and accuracy when compared with other approaches that employ alternative strategies.

In this paper we discuss the concept of co-adaptation between a human operator and a machine interface and we summarize its application with emphasis on two different domains, teleoperation and assistive technology. The analysis of the literature reveals that only in few cases the possibility of a temporal evolution of the co-adaptation parameters has been considered. In particular, it has been overlooked the role of time-related indexes that capture changes in motor and cognitive abilities of the human operator. We argue that for a more effective long-term co-adaptation process, the interface should be able to predict and adjust its parameters according to the evolution of human skills and performance. We thus propose a novel approach termed progressive co-adaptation, whereby human performance is continuously monitored and the system makes inferences about changes in the users' cognitive and motor skills. We illustrate the features of progressive co-adaptation in two possible applications, robotic telemanipulation and active vision for the visually impaired.

Spatial interactions between agents (humans, animals, or machines) carry information of high value to human or electronic observers. However, not all the information contained in a pair of continuous trajectories is important and thus the need for qualitative descriptions of interaction trajectories arises. The Qualitative Trajectory Calculus (QTC) (Van de Weghe, 2004) is a promising development towards this goal. Numerous variants of QTC have been proposed in the past and QTC has been applied towards analyzing various interaction domains. However, an inherent limitation of those QTC variations that deal with lateral movements is that they are limited to two-dimensional motion; therefore, complex three-dimensional interactions, such as those occurring between flying planes or birds, cannot be captured. Towards that purpose, in this paper QTC3D is presented: a novel qualitative trajectory calculus that can deal with full three-dimensional interactions. QTC3D is based on transformations of the Frenet–Serret frames accompanying the trajectories of the moving objects. Apart from the theoretical exposition, including definition and properties, as well as computational aspects, we also present an application of QTC3D towards modeling bird flight. Thus, the power of QTC is now extended to the full dimensionality of physical space, enabling succinct yet rich representations of spatial interactions between agents.

In this paper we propose to augment a wellestablished Qualitative Trajectory Calculus (QTC) by incorporating social distances into the model to facilitate a richer and more powerful representation of Human-Robot Spatial Interaction (HRSI). By combining two variants of QTC that implement different resolutions and switching between them based on distance thresholds we show that we are able to both reduce the complexity of the representation and at the same time enrich QTC with one of the core HRSI concepts: proxemics. Building on this novel integrated QTC model, we propose to represent the joint spatial behaviour of a human and a robot employing a probabilistic representation based on Hidden Markov Models. We show the appropriateness of our approach by encoding different HRSI behaviours observed in a human-robot interaction study and show how the models can be used to represent and classify these behaviours using social distance-augmented QTC.

We describe an integrated, real-time multi-camera surveillance system that is able to find and track individuals, acquire and archive facial image sequences, and perform face recognition. The system is based around an inference engine that can extract high-level information from an observed scene, and generate appropriate commands for a set of pan-tilt-zoom (PTZ) cameras. The incorporation of a reliable facial recognition into the high-level feedback is a main novelty of our work, showing how high-level understanding of a scene can be used to deploy PTZ sensing resources effectively. The system comprises a distributed camera system using SQL tables as virtual communication channels, Situation Graph Trees for knowledge representation, inference and high-level camera control, and a variety of visual processing algorithms including an on-line acquisition of facial images, and on-line recognition of faces by comparing image sets using subspace distance. We provide an extensive evaluation of this method using our system for both acquisition of training data, and later recognition. A set of experiments in a surveillance scenario show the effectiveness of our approach and its potential for real applications of cognitive vision.

The analysis and understanding of human-robot joint spatial behaviour (JSB) - such as guiding, approaching, departing, or coordinating movements in narrow spaces - and its communicative and dynamic aspects are key requirements on the road towards more intuitive interaction, safe encounter, and appealing living with mobile robots. This endeavours demand for appropriate models and methodologies to represent JSB and facilitate its analysis. In this paper, we adopt a qualitative trajectory calculus (QTC) as a formal foundation for the analysis and representation of such spatial behaviour of a human and a robot based on a compact encoding of the relative trajectories of two interacting agents in a sequential model. We present this QTC together with a distance measure and a probabilistic behaviour model and outline its usage in an actual JSB study. We argue that the proposed QTC coding scheme and derived methodologies for analysis and modelling are flexible and extensible to be adapted for a variety of other scenarios and studies.

One of fundamental issues for service robots is human-robot interaction. In order to perform such a task and provide the desired services, these robots need to detect and track people in the surroundings. In this paper, we propose a solution for human tracking with a mobile robot that implements multisensor data fusion techniques. The system utilizes a new algorithm for laser-based leg detection using the onboard laser range finder (LRF). The approach is based on the recognition of typical leg patterns extracted from laser scans, which are shown to also be very discriminative in cluttered environments. These patterns can be used to localize both static and walking persons, even when the robot moves. Furthermore, faces are detected using the robot's camera, and the information is fused to the legs' position using a sequential implementation of unscented Kalman filter. The proposed solution is feasible for service robots with a similar device configuration and has been successfully implemented on two different mobile platforms. Several experiments illustrate the effectiveness of our approach, showing that robust human tracking can be performed within complex indoor environments.