In recent years, the use of multirotor unmanned aerial vehicles (UAVs) has expanded significantly across various fields. Concurrently, aerial manipulation has emerged as a thriving area, leading to the development of diverse aerial manipulators [1]. The most common method is the integration of arms into multirotors [2–4]. Utilizing rotor-concentrated manipulators (RCMs), these drones can perform various manipulation tasks [5–7].
We present here proof of concept for a smart energy-efficient hybrid gait monitoring system. It consists of a hybrid Inertial Measuring Unit (IMU) and Triboelectric Nanogenerator (TENG) sensing module, and it utilizes multiple strategies to reduce the computational load and energy overhead of the device while maintaining accuracy. It meets the lower energy demands through a self-powered mechanism, thus eliminating the need for batteries. Firstly, a TENG harvests the biomechanical energy of human movement and is optimized for high energy output. Then the IMU sensor collects and processes only sparse data. An algorithm is implemented to select the best sampling points in the gait data, and it adapts as it gains knowledge of the user's walking pattern. The IMU sensor is powered intermittently through the TENG and a power management unit while the person is walking. Finally, a TinyML algorithm is employed for data intelligence at the device level.
This paper details the development, implementation and experimental evaluation of an interface-aware, task-agnostic assistance system for shared human-robot teleoperation, specifically applied to a 7-DoF robotic arm. The system addresses a limitation of current shared-control methods by considering the impact of control interfaces on user input precision and the robot agent's incomplete understanding of the human's policy. The approach is evaluated in empirical case studies involving participants with spinal cord injuries. The personalized assistance system improves safety and reduces cognitive load.
In this paper, we investigate the strategies used by a skilled ultrasound physician and explore the functionality of a robotic ultrasound scan acquisition system. We develop a framework to record an expert radiologist's probe motion, probe force, audio narration, and generated ultrasound images. Then, we optimize the acquired scan data to remove redundant sections and use the processed data to perform an autonomous survey scan with the robotic system. Finally, the radiologist performs a full detailed scan of the abdomen through a haptic user interface capable of long-distance remote interaction. We compare images generated at similar points in each scan and identify the same anatomy and pathologies, thus demonstrating equivalent diagnostic performance between the skilled ultrasound physician and robotic system.
This paper presents a distributed, cooperative persistent coverage control for multi-robotic systems with limited communication and limited sensing range. The control framework synthesizes computational geometry-based algorithms with an awareness-based model to incentivize the frequent coverage of every point of interest in a large-scale domain. Theoretical results on the stability and convergence of awareness are given. The performance of the control approach is demonstrated via an experiment with a hexapod in a simulated cooperative environment.
This paper presents a new assembler robot intended as part of a larger modular self-reconfigurable industrial machine system, with a design focus on (1) the integration of proprioceptive actuators into the assembler robot, and (2) the implementation of kinematic couplings in the modular connectors. We propose two comparison metrics concerning energy efficiency and mechanical repeatability, anticipating their growing importance as robot-assembled modular systems expand in scale. Our prototype assembler robot, Belty, employs high torque density quasi-direct drive belt transmission BLDC actuators. These offer benefits including superior efficiency, back-drivability, impact resilience, and the ability to perform dynamic motions such as hopping. Robust dynamic motion is advantageous for assembly tasks in restricted spaces, where contact-rich actions, such as dragging or bumping, can streamline assembly operations and algorithms. Additionally, we delve into the mechanical design of our connector, which is based on an exact-constraint mechanism called a kinematic coupling. This connector achieves a wide area of acceptance while also offering outstanding repeatability on the order of low tens of micrometers (10 m ).
Recent work in Task and Motion Planning (TAMP) has enabled a new class of algorithms that can better take advantage of off-the-shelf samplers and solvers to find solutions to sub-problems in a task plan, such as motion between configurations, or inverse kinematics solutions. However, not all sub-problems are equally valuable. Existing planners typically rely on heuristics to determine which sub-problem to attempt to solve next, unable to reason about the expected cost of doing so in the broader context of the full plan. In this work, we present a novel approach for TAMP, utilizing learned models to inform when to attempt to solve potentially expensive sub-problems. We test our approach in two simulated domains, as well as on a real Panda robot, showing improvement in planning and execution time compared to a heuristic driven baseline.
This work addresses the challenge of fault recovery from limb-loss by employing novel dynamical gaits. We demonstrate empirically a stable tripedal dynamic gait on the Ghost Robotics Minitaur platform. This pronking gait achieves comparable speeds to those on the intact quadrupedal machine. In consequence, despite substantially higher energetic expenditure per stride, it’s CoT is reduced by ∼ 10
We introduce an autonomous driving framework that employs convolutional neural networks. This framework utilizes forward-facing stereo camera images, vehicle speed, traffic light status, and higher-level navigation commands to predict future waypoints for the vehicle's trajectory. The model was trained on a dataset collected from the CARLA Simulator and underwent testing in both simulation and real-world settings without any additional fine-tuning on real-world datasets. In simulation testing, the model successfully navigated previously unseen maps and weather conditions, covering a distance of 3000m without encountering collisions or traffic light violations. Real-world testing on a differential drive vehicle demonstrated the model's ability to navigate without lane invasions.
Notwithstanding recent progress in the control of prosthetic limbs, the independent control of a multiplicity of joints, such as in pluriarticulated proximal upper-limb amputations, remains complex and tiresome to users. As an alternative or an addition to classical myoelectric control, few studies have indicated the possibility of measuring compensatory motions of residual limbs of users for controlling their prostheses. In this paper, we introduce an algorithm that generalizes this idea, allows to interpret residual limb motions as encoding the user's intention of motion, and translate it into prosthesis control to achieve a desired reaching task. We validate this approach in simulation for a transhumeral subject, and experimentally demonstrating the control of a 5-DoF robotic arm via human motions tracked by Inertial Measurement Units (IMUs) sensors.
The Tethered Aircraft Unmanned System enables long-term distributed sampling at altitude via an array of sensor suites embedded along the power tether. As the aircraft traverses the location of the sensors along the tether and during operation remains unknown. In this work, we propose an inertial-based method for estimating real-time localization of the sensor array. We have integrated an inertial measurement unit into each sensor suite and the orientation of each sensor is transformed into the world frame using a Madgwick estimator. The power tether is modeled as an asymmetric catenary which enables us to define a subspace of the explorable volume for system initialization and periodic online re-calibrations. The system flew numerous curvilinear trajectories in a laboratory setting. Analysis of the intercomparison between our approach and Vicon ground truth positional data shows average RMSE of 0.524, 0.493, and 0.625m in the N, E, and D axes after 10 s, respectively.
Soft actuators have garnered substantial interest in robotics, promising safer and more adaptable human-robot interaction. This paper presents the development and evaluation of a novel electromagnetic soft actuator, named “MagFlex”, designed to address critical challenges associated with existing soft actuators. MagFlex demonstrates remarkable capabilities, with a load bearing capacity of 650 g and contraction strain of 33
Bio-acoustic monitoring (BAM) is the science of studying fauna and the environment through audio monitoring, recording, and analysis. This is a newly developing field with growing interest due to rapid changes to animal habitats and environments due to global warming (Welz, 2019).
This paper presents a novel model-free method for humanoid robots to self-calibrate their foot force sensors using quasi-static movement control. The robot generates whole-body trajectories using our model-free framework with three steps: 1. Designing target trajectories containing discrete pairs of the center of pressure (CoP) and foot position. 2. Formulating a cost function that minimizes the error between the measured and desired objectives. 3. Updating the robot motion with an optimization algorithm by slightly moving the robot's leg joints back and forth. The foot sensor parameters are then determined by minimizing the error between the measured and reference CoP and ground reaction force (GRF) during the robot's movement. We demonstrate our approach on a NAO humanoid robot platform. Experimental results show that the model-free self-calibrated shoes can successfully estimate CoP and GRF.
Complex-shaped objects, e.g., tangled-prone or flexible objects, pose challenges in industrial bin picking. Due to the heavy occlusion and the elusive entanglement situation when these objects are randomly placed in a bin, it is difficult for the robot to pick only one object at a time. Previous works address this issue by abstracting away state estimation for the clutter and directly predicting the potentially entangled objects from visual input, making the entanglement estimation lack the proper geometrical information of the objects. In this paper, we propose a novel bin-picking system that can grasp avoiding entanglement for both rigid and flexible objects. Our method can (1) infer the full state of the objects in the clutter by skeletonizing and restoring the occluded shapes and (2) estimate the degree of entanglement quantitatively for each restored object and determine the grasping object based on the degree of entanglement, height information and the restoration rate. To evaluate our method, we perform real-world experiments using rigid objects with different shapes and one deformable object. Experimental results also demonstrate the effectiveness of our method on various objects with fixed or irregular geometries.
This work presents an energy-aware PRM*-based path planner, energy estimator, and landing control algorithm that will allow a low-cost fixed-wing UAV to carry out a combined mission of remote sensing while flying and direct water sampling at landing. We performed several experiments in land and water environments under sun, snow, rain, day, and evening conditions with our modified off-the-shelf RC aircraft. The aircraft accomplished the mission, flying a survey pattern, landing at a series of sampling points and then returning to the starting location while respecting the available energy budget. These innovations enable high-impact applications, such as environmental monitoring.
In autonomous driving perception system, 3D object detection task plays a crucial role. Recently, although many detection methods with camera only input have shown good performance by retrieving high resolution and rich semantic from images, they are still limited by the lack of depth information. On the other hand, automotive radar, as a common on-board sensor, is typically only employed to perform low-level perception tasks despite its ability to provide accurate depth and doppler velocity information. Therefore, effectively fusing camera and radar for detection can leverage the advantages of both sensors without incurring additional costs. In this paper, we propose a novel Attentive Radar-Camera fusion in Bird-Eye-View (ARC-BEV) 3d detection framework for autonomous driving. Unlike current radar-camera fusion methods, ARC-BEV transforms images into BEV features and fuses them with radar data, thereby mitigating issues of alignment loss and insufficient height information. A spatial attention fusion module is also incorporated to enhance the relevance of features in the spatial domain. ARC-BEV achieves state-of-the-art performance with 43.7
In the realm of traditional humanoid torque control, a prevalent practice involves meticulously tracking preplanned joint motions using high-frequency controllers to solve locomotion problems. However, recent studies have suggested the possibility of reducing excessive control frequencies of torque controllers through the utilization of deep reinforcement learning policies. This paper presents the impact of control frequency on torque-based deep reinforcement learning controllers, ranging from 250 Hz to 40 Hz. The study also outlines the method used to train the torque-based deep RL policy in various control frequencies while isolating the effect of frequency changes to ensure fair performance comparisons. The lower the control frequency, the more robust the results were against robot system delays, even on unexpected terrain or at higher target walking speeds within the training range.
Human-robot interaction is a critical area of research, providing support for collaborative tasks where a human instructs a robot to interact with and manipulate objects in an environment.
This paper presents AMoRPH, a new analytical inverse kinematics solver for extensible multi-section continuum robots. It uses a virtual linkage model to efficiently solve the 5 DoF inverse kinematics problem and find balanced, smooth solutions. Segment addition and geometry adjustments handle non-uniform cases and obstacles. Simulations and hardware experiments validate the method's speed, accuracy, and versatility. For a 100-step circular trajectory, AMoRPH computed solutions in milliseconds, compared to seconds for other algorithms. The balanced curvatures, precise tracking, and smooth motions demonstrate its capabilities. This enables utilizing continuum robots requiring millisecond response times for human interaction and assistance. The efficient formulation guarantees solutions and enables real-time control. Obstacle avoidance is also implemented based on collision detection with the virtual linkage model. Experiments validate the solver's real-time performance, accuracy, and smoothness on both simulated and physical extensible continuum robots.
