Frontiers in Robotics and AI最新文献

The subject of this article is the development of an unmanned surface vehicle (USV) for the removal of floating debris. A twin-hulled boat with four thrusters placed at the corners of the vessel is used for this purpose. The trash is collected in a storage space through a timing belt driven by an electric motor. The debris is accumulated in a funnel positioned at the front of the boat and subsequently raised through this belt into the garbage bin. The boat is equipped with a spherical camera, a long-range 2D LiDAR, and an inertial measurement unit (IMU) for simultaneous localization and mapping (SLAM). The floating debris is identified from rectified camera frames using YOLO, while the LiDAR and IMU concurrently provide the USV's odometry. Visual methods are utilized to determine the location of debris and obstacles in the 3D environment. The optimal order in which the debris is collected is determined by solving the orienteering problem, and the planar convex hull of the boat is combined with map and obstacle data via the Open Motion Planning Library (OMPL) to perform path planning. Pure pursuit is used to generate the trajectory from the obtained path. Limits on the linear and angular velocities are experimentally estimated, and a PID controller is tuned to improve path following. The USV is evaluated in an indoor swimming pool containing static obstacles and floating debris.

Tendon-driven continuum actuators (TDCAs) provide compliant and lifelike motion that is well suited for human-robot interaction, but their structural compliance and underactuation make them susceptible to undesired vibrations, particularly along unactuated axes under load. This work addresses vibration suppression in such systems by proposing a real-time control strategy for a two-degree-of-freedom TDCA-based soft robotic neck used in the HARU social robot, where yaw motion is unactuated and prone to oscillations due to eccentric loading. The proposed approach combines a current-based tendon pretensioning routine, baseline PID control of the actuated pitch and roll axes, and a novel Coupled Axis Indirect Vibration Suppression (CIVS) mechanism. CIVS exploits mechanical cross-axis coupling by using high-pass filtered yaw acceleration from an inertial sensor to generate transient tension modulations in the actuated tendons, thereby increasing effective damping of the unactuated yaw mode without introducing additional hardware or compromising compliance. A classical sliding mode control is also implemented as a nonlinear benchmark under identical hardware constraints. Experimental validation on the HARU neck under representative loading conditions demonstrates that the proposed method achieves substantial vibration attenuation. Compared to the baseline controller, CIVS reduces yaw angular range by approximately 53% and yaw acceleration area by over 60%, while preserving smooth, expressive motion. The results further show that CIVS outperforms the sliding mode controller in suppressing vibrations on the unactuated axis. These findings indicate that indirect, feedback-driven tendon modulation provides an effective and low-complexity solution for mitigating load-induced vibrations in underactuated soft robotic systems, making the approach particularly suitable for interactive applications where safety, compliance, and motion expressivity are critical.

Factor graph optimization serves as a fundamental framework for robotic perception, enabling applications such as pose estimation, simultaneous localization and mapping (SLAM), structure-from-motion (SfM), and situational modeling. Traditionally, these methods solve unconstrained least squares problems using algorithms such as Gauss-Newton and Levenberg-Marquardt. However, extending factor graphs with native support for hard equality constraints can yield more accurate state estimates and broaden their applicability, particularly in planning and control. Prior work has addressed equality handling either by soft penalties (large weights) or by nested-loop Augmented Lagrangian (AL) schemes. In this paper, we propose a novel extension of factor graphs that seamlessly incorporates hard equality constraints without requiring additional optimization techniques. Our approach maintains the efficiency and flexibility of existing second-order optimization techniques while ensuring constraint satisfaction. To validate the proposed method, an autonomous-vehicle velocity-tracking optimal control problem is solved and benchmarked against an AL baseline, both implemented in g2o. Additional comparisons are conducted in GTSAM, where the penalty method and AL are evaluated against our g2o implementations. Moreover, we introduce ecg2o, a header-only C++ library that extends the widely used g2o library with full support for hard equality-constrained optimization. This library, along with demonstrative examples and the optimal control problem, is available as open source at https://github.com/snt-arg/ecg2o.

Gaze is a fundamental aspect of non-verbal communication in human interaction, playing an important role in conveying attention, intentions, and emotions. A key concept in gaze-based human interaction is joint attention, the focus of two individuals on an object in a shared environment. In the context of human-robot interaction (HRI), gaze-following has become a growing research area, as it enables robots to appear more socially intelligent, engaging, and likable. While various technical approaches have been developed to achieve this capability, a comprehensive overview of existing implementations has been lacking. This scoping review addresses this gap by systematically categorizing existing solutions, offering a structured perspective on how gaze-following behavior is technically realized in the field of HRI. A systematic search was conducted across four databases, leading to the identification of 28 studies. To structure the findings, a taxonomy was developed that categorizes technological approaches along three key functional dimensions: (1) environment tracking, which involves recognizing the objects in the robot's surroundings; (2) gaze tracking, which refers to detecting and interpreting human gaze direction; and (3) gaze-environment mapping, which connects gaze information with objects in the shared environment to enable appropriate robotic responses. Across studies, a distinction emerges between constrained and unconstrained solutions. While constrained approaches, such as predefined object positions, provide high accuracy, they are often limited to controlled settings. In contrast, unconstrained methods offer greater flexibility but pose significant technical challenges. The complexity of the implementations also varies significantly, from simple rule-based approaches to advanced, adaptive systems that integrate multiple data sources. These findings highlight ongoing challenges in achieving robust and real-time gaze-following in robots, particularly in dynamic, real-world environments. Future research should focus on refining unconstrained tracking methods and leveraging advances in machine learning and computer vision to make human-robot interactions more natural and socially intuitive.

The ability to track the 6D pose distribution of an object while a mobile manipulator is still approaching it can enable the robot to pre-plan grasps, thereby improving both the time efficiency and robustness of mobile manipulation. However, tracking a 6D object pose distribution on approach can be challenging due to the limited view of the robot camera. In this study, we present a particle filter-based multi-view 6D pose distribution tracking framework that compensates for the limited view of the moving robot camera while it approaches the object by fusing observations from external stationary cameras in the environment. We extend the single-view pose distribution tracking framework (PoseRBPF) to fuse observations from external cameras. We model the object pose posterior as a multi-modal distribution and introduce techniques for fusion, re-sampling, and pose estimation from the tracked distribution to effectively handle noisy and conflicting observations from different cameras. To evaluate our framework, we also contribute a real-world benchmark dataset. Our experiments demonstrate that the proposed framework yields a more accurate quantification of object pose and associated uncertainty than previous research. Finally, we apply our framework for pre-grasp planning on mobile robots, demonstrating its practical utility.

This paper explores the integration of Physics-Informed Neural Networks (PINNs) and Robot Process Automation (RPA) tools in modeling and controlling rigid robotic joint motion. PINNs, which integrate physical laws with neural networks, offer a promising solution for solving both forward and inverse problems in robotics, while RPA tools provide the means to automate and streamline these processes. The study discusses various PINN techniques, including Extended PINNs, Hybrid PINNs, and Minimized Loss techniques, developed to address issues such as high training costs and slow convergence rates. By combining these advanced PINN approaches with RPA tools, the research aims to enhance the precision and efficiency of robot control, motion planning, and process automation, particularly in non-linear and dynamic coupling situations. We also examine PDE-Inspired PINNs for motion planning in robot navigation and manipulation by integrating it with ROS using the RPA tool itself for coordinating joints and angle movements, and exploring how RPA can facilitate the implementation of these models in real-world scenarios.

Although cognitive robotics is still a work in progress, the trend is to "free" robots from the assembly lines of the third industrial revolution and allow them to "enter human society" in large numbers and many forms, as forecasted by Industry 4.0 and beyond. Cognitive robots are expected to be intelligent, designed to learn from experience and adapt to real-world situations rather than being preprogrammed with specific actions for all possible stimuli and environmental conditions. Moreover, such robots are supposed to interact closely with human partners, cooperating with them, and this implies that robot cognition must incorporate, in a deep sense, ethical principles and evolve, in conflict situations, decision-making capabilities that can be perceived as wise. Intelligence (true vs. false), ethics (right vs. wrong), and wisdom (good vs. bad) are interrelated but independent features of human behavior, and a similar framework should also characterize the behavior of cognitive agents integrated in human society. The working hypothesis formulated in this paper is that the propensity to consolidate ethically guided behavior, possibly evolving to some kind of wisdom, is a cognitive architecture based on bio-inspired embodied cognition, educated through development and social interaction. In contrast, the problem with current AI foundation models applied to robotics (EAI) is that, although they can be super-intelligent, they are intrinsically disembodied and ethically agnostic, independent of how much information was absorbed during training. We suggest that the proposed alternative may facilitate social acceptance and thus make such robots civilized.

Cybernetic avatars are hybrid interaction robots or digital representations that combine autonomous capabilities with teleoperated control. This study investigates the acceptance of cybernetic avatars, with particular emphasis on robot avatars for customer service. Specifically, we explore how acceptance varies as a function of modality (physical vs. virtual), robot appearance (e.g., android, robotic-looking, cartoonish), deployment settings (e.g., shopping malls, hotels, hospitals), and functional tasks (e.g., providing information, patrolling). To this end, we conducted a large-scale survey with over 1,000 participants in Dubai. As one of the most multicultural societies worldwide, Dubai offers a rare opportunity to capture opinions from multiple cultural clusters within a single setting simultaneously, thereby overcoming the limitations of nationally bound samples and providing a more global picture of acceptance. Overall, cybernetic avatars received a high level of acceptance, with physical robot avatars receiving higher acceptance than digital avatars. In terms of appearance, robot avatars with a highly anthropomorphic robotic appearance were the most accepted, followed by cartoonish designs and androids. Animal-like appearances received the lowest level of acceptance. Among the tasks, providing information and guidance was rated as the most valued. Shopping malls, airports, public transport stations, and museums were the settings with the highest acceptance, whereas healthcare-related spaces received lower levels of support. An analysis by community cluster revealed, among other findings, that Emirati respondents were particularly accepting of android appearances, whereas participants from the 'Other Asia' cluster were particularly accepting of cartoonish appearances. Our study underscores the importance of incorporating citizen feedback from the early stages of design and deployment to enhance societal acceptance of cybernetic avatars.

This paper proposes an exploration-efficient deep reinforcement learning with reference (DRLR) policy framework for learning robotics tasks incorporating demonstrations. The DRLR framework is developed based on an imitation bootstrapped reinforcement learning (IBRL) algorithm. Here, we propose to improve IBRL by modifying the action selection module. The proposed action selection module provides a calibrated Q-value, which mitigates the bootstrapping error that otherwise leads to inefficient exploration. Furthermore, to prevent the reinforcement learning (RL) policy from converging to a sub-optimal policy, soft actor-critic (SAC) is used as the RL policy instead of twin delayed DDPG (TD3). The effectiveness of our method in mitigating the bootstrapping error and preventing overfitting is empirically validated by learning two robotics tasks: bucket loading and open drawer, which require extensive interactions with the environment. Simulation results also demonstrate the robustness of the DRLR framework across tasks with both low and high state-action dimensions and varying demonstration qualities. To evaluate the developed framework on a real-world industrial robotics task, the bucket loading task is deployed on a real wheel loader. The sim-to-real results validate the successful deployment of the DRLR framework.

Spoken language is one of the most powerful tools for humans to learn, exchange information, and build social relationships. An inherent feature of spoken language is large within- and between-speaker variation across linguistic levels, from sound acoustics to prosodic, lexical, syntactic, and pragmatic choices that differ from written language. Despite advancements in text-to-speech and language models used in social robots, synthetic speech lacks human-like variability. This limitation is especially critical in interactions with children, whose developmental needs require adaptive speech input and ethically responsible design. In child-robot interaction research, robot speech design has received less attention than appearance or multimodal features. We argue that speech variability in robots needs closer examination, considering both how humans adapt to robot speech and how robots could adjust to human speech. We discuss three tensions: (1) feasibility, because dynamic human speech variability is technically challenging to model; (2) desirability, because variability may both enhance and hinder learning, usability, and trust; and (3) ethics, because digital human-like speech risks deception, while robot speech varieties may support transparency. We suggest approaching variability as a design tool while being transparent about the robot's role and capabilities. The key question is which types of variation benefit children's socio-cognitive and language learning, at which developmental stage, in which context, depending on the robot's role and persona. Integrating insights across disciplines, we outline directions for studying how specific dimensions of variability affect comprehension, engagement, language learning, and for developing vocal interactivity that is engaging, ethically transparent, and developmentally appropriate.