Soft robotic arms based on flexible materials or structures have gained popularity in recent years due to their superior adaptability and safety. However, controlling soft arms faces challenges due to various nonlinear problems, such as vis-coelastic effects. The viscoelastic effects can be categorized into dynamic viscoelasticity, which occurs in short term (e.g. about 10 minutes), and steady-state viscoelasticity, which manifests over longer periods (e.g. more than one hour). Although a small number of studies have explored the viscoelastic properties of soft robots before, there is still limited research on the control of soft arms taking into account viscoelastic effects, especially the steady-state viscoelastic effects. This paper proposes a method that combines modeling and learning to address the impact of dynamic and steady-state viscoelasticity on soft robot control. This paper introduces a method for separating between dynamic and steady-state viscoelastic effects and establishes a model for steady-state viscoelasticity. By integrating this model with a recurrent neural network (RNN), we achieve stable and accurate open-loop control of the flexible arm. Experimental results demonstrate that the RNN effectively compensates for the influence of historical motion states on dynamic viscoelasticity. Moreover, the incorporation of the steady-state viscoelastic model enables the controller to maintain high accuracy in open-loop control experiments for up to five hours.

Soft robotic arms have shown great potential toward applications to human daily lives, which is mainly due to their infinite passive degrees of freedom and intrinsic safety. There are tasks in lives that require the motion of the robot to meet some certain pose constraints that have not been implemented through the soft arm, like delivering a glass of water. Because the workspace of the soft arm is affected by the loads or interaction, it is difficult to implement this task through the motion planning method. In this letter, we propose a Q-learning based approach to address the problem, directly achieving motion control with constraints under loads and interaction without planning. We first generate a controller for the soft arm based on Q-learning, which can operate the arm while satisfying the pose constraints when the arm is neither loaded nor interacted with the environment. Then, we introduce a process that adjusts corresponding Q values in the controller, which allows the controller to operate the arm with an unknown load or interaction while still satisfying the pose constraints. We implement the approach on our soft arm, i.e., the Honeycomb Pneumatic Network (HPN) Arm. The experiments show that the approach is effective, even when the arm reached an untrained situation or even beyond the workspace under the interaction.

It is challenging to control a soft robot, where reinforcement learning methods have been applied with promising results. However, due to the poor sample efficiency, reinforcement learning methods require a large collection of training data, which limits their applications. In this paper, we propose a Q-learning controller for a physical soft robot, in which pre-trained models using data from a rough simulator are applied to improve the performance of the controller. We implement the method on our soft robot, i.e., Honeycomb Pneumatic Network (HPN) arm. The experiments show that the usage of pre-trained models can not only reduce the amount of the real-world training data, but also greatly improve its accuracy and convergence rate.

This paper presents an adaptive, simple, and effective guidance approach for hypersonic entry vehicles with high lift-to-drag (L/D) ratios (e.g., hypersonic gliding vehicles). The core of the constrained guidance approach is a closed-form, easily obtained, and computationally efficient feedback control law that yields the analytic bank command based on the well-known quasi-equilibrium glide condition (QEGC). The magnitude of the bank angle command consists of two parts, i.e., the baseline part and the augmented part, which are calculated analytically and successively. The baseline command is derived from the analytic relation between the range-to-go and the velocity to guarantee the range requirement. Then, the bank angle is augmented with the predictive altitude-rate feedback compensations that are represented by an analytic set of flight path angle needed for the terminal constraints. The inequality path constraints in the velocity-altitude space are translated into the velocity-dependent bounds for the magnitude of the bank angle based on the QEGC. The sign of the bank command is also analytically determined using an automated bank-reversal logic based on the dynamic adjustment criteria. Finally, a feasible three-degree-of-freedom (3DOF) entry flight trajectory is simultaneously generated by integrating with the real-time updated command. Because no iterations and no or few off-line parameter adjustments are required using almost all analytic processing, the algorithm provides remarkable simplicity, rapidity, and adaptability. A considerable range of entry flights using the vehicle data of the CAV-H is tested. Simulation results demonstrate the effectiveness and performance of the presented approach.

This paper presents the design, control, and applications of a multi-segment soft robotic arm. In order to design a soft arm with large load capacity, several design principles are proposed by analyzing two kinds of buckling issues, under which we present a novel structure named Honeycomb Pneumatic Networks (HPN). Parameter optimization method, based on finite element method (FEM), is proposed to optimize HPN Arm design parameters. Through a quick fabrication process, several prototypes with different performance are made, one of which can achieve the transverse load capacity of 3 kg under 3 bar pressure. Next, considering different internal and external conditions, we develop three controllers according to different model precision. Specifically, based on accurate model, an open-loop controller is realized by combining piece-wise constant curvature (PCC) modeling method and machine learning method. Based on inaccurate model, a feedback controller, using estimated Jacobian, is realized in 3D space. A model-free controller, using reinforcement learning to learn a control policy rather than a model, is realized in 2D plane, with minimal training data. Then, these three control methods are compared on a same experiment platform to explore the applicability of different methods under different conditions. Lastly, we figure out that soft arm can greatly simplify the perception, planning, and control of interaction tasks through its compliance, which is its main advantage over the rigid arm. Through plentiful experiments in three interaction application scenarios, human-robot interaction, free space interaction task, and confined space interaction task, we demonstrate the potential application prospect of the soft arm.

We present a hand specialized for climbing unstructured rocky surfaces. Articulated fingers achieve grasps commonly used by human climbers. The gripping surfaces are equipped with dense arrays of spines that engage with asperities on hard rough materials. A load-sharing transmission system divides the shear contact force among spine tiles on each phalanx to prevent premature spine slippage or grasp failure. Taking advantage of the hand’s kinematic and load-sharing properties, the wrench space of achievable forces and moments can be computed rapidly. Bench-top tests show agreement with the model, with average wrench space errors of 10–15%, despite the stochastic nature of spine/surface interaction. The model provides design guidelines and control strategy insights for the SpinyHand and can inform future work.

Soft variable-length continuum manipulators have emerged as ideal agents in common human interaction scenarios owing to their flexible movements, large workspace and safety assurance. Besides, it’s also their variable-length property that leads to a much larger configuration space, which makes it more difficult to solve the motion planning problem using state-of-the-art sampling-based methods. In this paper, we propose an algorithm that fully exploits the variable-length property of these manipulators. The algorithm directly generates and selects feasible configuration nodes in task space. Before that, a path of the manipulator’s end effector is pre-generated in task space according to the positions of the goal and obstacles, which provides approximately accurate guiding direction of the whole manipulator. During the simulation experiments, the two-level algorithm is validated with efficiency in comparison of Jacobian-based methods and other extensional tasks.

In this paper, aiming at fully taking advantage of soft manipulators working ability, we propose a parameterized approximating method to characterize their force output in full workspace. We define the Workspace-Load bearing capacity Cloud (WLC) of soft arms and present the method to calculate WLC in the three-dimensional space by linear fitting. At last, finite element analysis is used to validate its effectiveness in characterizing the force output of soft arms in full workspace.

Grasping objects that are too large to envelop is traditionally achieved using friction that is activated by squeezing. We present a family of shear-activated grippers that can grasp such objects without the need to squeeze. When a shear force is applied to the gecko-inspired material in our grippers, adhesion is turned on; this adhesion in turn results in adhesion-controlled friction, a friction force that depends on adhesion rather than a squeezing normal force. Removal of the shear force eliminates adhesion, allowing easy release of an object. A compliant shear-activated gripper without active sensing and control can use the same light touch to lift objects that are soft, brittle, fragile, light, or very heavy. We present three grippers, the first two designed for curved objects, and the third for nearly any shape. Simple models describe the grasping process, and empirical results verify the models. The grippers are demonstrated on objects with a variety of shapes, materials, sizes, and weights.

In view of the inflexibility of human-computer interaction for the robot to help elder and disabled, this paper introduces the recognition technology of facial expression to integrate with the robot. So, the robot can perform corresponding actions according to different facial expression, which greatly improves the convenience for human to control robot. This thesis mainly focuses on the accuracy of face location, the recognition rate of expression recognition and the application of expression recognition system. This thesis is trained and tested on the HELEN face dataset and the self-built face dataset respectively. The tests achieve good results in both face alignment experiments and facial expression recognition experiments. Finally, by combining the expression recognition system with robot platform of helping elder and disabled, the purpose of controlling the movement of the robot by expression information is realized, which proves the feasibility of the method.

Shaft sinking by drilling is widely used in mining, construction and coal industry. Measuring the shaft borehole accurately, is of great importance to the quality in these projects among other factors. This paper proposes a novel design and implementation of ultrasonic logging instrument for shaft sinking by drilling (ULISSD). Design of the ultrasonic ranging module, depth module and the module for downhole orientation, as well as the borehole radius measurement algorithm are explained and illustrated. This paper also demonstrates an experimental application of ULISSD, which shows that, compared with other similar instruments, ULISSD is able to measure accurately borehole shaft radius wider diameters (3-12 m) at deeper vertical sections (maximum 1000 m) in slurry with higher gravity (<= 1.22). (C) 2018 Published by Elsevier Ltd.

This paper presents models of arrays of compliantly supported spines that attach to rough surfaces. The applications include climbing and perching robots. Surfaces are characterized in terms of asperity distributions, which lead to stochastic models of spine force capabilities over a range of loading directions. Models cover unidirectional spine arrays and pairs of opposed arrays that withstand normal forces pulling away from a surface. Experiments on a variety of surfaces confirm the predicted behavior. For opposed spine arrays, the overall load capability also depends on the preloading strategy for applying internal forces. Insights from the analysis guide the design of spine array mechanisms to allow, for example, a small aerial platform to attach to walls and ceilings.

Grasping and manipulating uncooperative objects in space is an emerging challenge for robotic systems. Many traditional robotic grasping techniques used on Earth are infeasible in space. Vacuum grippers require an atmosphere, sticky attachments fail in the harsh environment of space, and handlike opposed grippers are not suited for large, smooth space debris. We present a robotic gripper that can gently grasp, manipulate, and release both flat and curved uncooperative objects as large as a meter in diameter while in microgravity. This is enabled by (i) space-qualified gecko-inspired dry adhesives that are selectively turned on and off by the application of shear forces, (ii) a load-sharing system that scales small patches of these adhesives to large areas, and (iii) a nonlinear passive wrist that is stiff during manipulation yet compliant when overloaded. We also introduce and experimentally verify a model for determining the force and moment limits of such an adhesive system. Tests in microgravity show that robotic grippers based on dry adhesion are a viable option for eliminating space debris in low Earth orbit and for enhancing missions in space.

Most control methods of soft manipulators are developed based on physical models derived from mathematical analysis or learning methods. However, due to internal nonlinearity and external uncertain disturbances, it is difficult to build an accurate model, further, these methods lack robustness and portability among different prototypes. In this work, we propose a model-free control method based on reinforcement learning and implement it on a multi-segment soft manipulator in 2D plane, which focuses on the learning of control strategy rather than the physical model. The control strategy is validated to be effective and robust in prototype experiments, where we design a simulation method to speed up the training process.

Free-flying robots have the potential to autonomously fulfill a wide range of tasks involving manipulation of objects in space. In this paper we study the design of a wrist mechanism for free-flying robots that are equipped with an adhesive gripper for attaching to objects and surfaces. The wrist and gripper allow the robots to apply moments in addition to forces, which increases their versatility for object manipulation. We apply grasp optimization to establish limitations on the forces/moments that the wrist can impart, subject to adhesion capabilities. Building on these results, we present considerations for tuning a passive wrist mechanism, or controlling an active wrist, to broaden the range of forces and moments that the robot can exert. Our theoretical insights and wrist designs are validated in simulations and on a planar micro-gravity test bed.

Perching and climbing as animals do is useful to aerial robots for extending mission life and for interacting with the physical world because flight is energetically costly. This paper presents the design and modeling of a claw or spine based gripper for perching on rough, curved surfaces. Drawing inspiration from the opposed grip techniques found in animals, we focus on the design considerations associated with surface geometry and preload. A model elucidates the relationship between these variables, and a mechanism demonstrates the effectiveness of the opposed grip technique.

Soft manipulators have been a rising focus of soft robotics research. Taking advantage of soft materials and flexible, continuous movements, they have promising applicable prospect. However, their highly internal nonlinearity and unpredictable deformation caused by environmental effects make it difficult to build an exact model for control. In this work, we propose a generalized controller for soft manipulators using an estimated Jacobian-based model derived from structural analysis. The model can be simplified from reasonable assumptions of manipulator structure, and updated to balance conformity to reality and stability. In prototype experiments on an 3D multi-segment soft manipulator, the control method exhibits accuracy as well as adaptability to self gravity and external loads.

Perching can extend the useful mission life of a micro air vehicle. Once perched, climbing allows it to reposition precisely, with low power draw and without regard for weather conditions. We present the Stanford Climbing and Aerial Maneuvering Platform, which is to our knowledge the first robot capable of flying, perching with passive technology on outdoor surfaces, climbing, and taking off again. We present the mechanical design and the new perching, climbing, and takeoff strategies that allow us to perform these tasks on surfaces such as concrete and stucco, without the aid of a motion capture system or off-board computation. We further discuss two new capabilities uniquely available to a hybrid aerial-scansorial robot: the ability to recover gracefully from climbing failures and the ability to increase usable foothold density through the application of aerodynamic forces. We alsomeasure real power consumption for climbing, flying, and monitoring and discuss how future platforms could be improved for longer mission life.

We present work on incipient slip sensing and recovery for controllable gecko-inspired adhesives. The approach is based on the relationship between changes in real contact area and maximum shear force. Using signals from an on-board tactile sensor, we detect the onset of adhesive failure and execute recovery behavior. Results show the system using tactile sensor feedback is able to achieve >92% of the peak adhesion performance achieved with a force plate and commercial load cell. Results are consistent over a variety of common smooth surfaces, with the system achieving repeatable force loading behavior independent of varying materials and surface conditions.

We present a new spine solution for the locomotion of human-scale robots on steep, rocky surfaces, known as linearly-constrained spines. The spine stiffness is low in the normal direction but high with respect to lateral and bending loads. The solution differs from previous spine arrays used for small robots in having a much higher spine density and less spine scraping over asperities. We present theoretical and empirical results to demonstrate that this solution is capable of shear stresses of over 200kPa, enabling human-scale robots to apply forces parallel to steep rock surfaces for climbing, bracing, etc. The analysis includes the effects of spine geometry, stiffness, backlash and three-dimensional loading angle to predict the overall forces possible in three dimensions of both single and opposed configurations of spine arrays. Demonstrated applications include a gripper for a “smart staff” aimed at helping humanoid robots to negotiate steep terrain and a palm that provides over 700N in shear for the RoboSimian quadruped.