Design and experiments on an inflatable link robot with a built-in vision sensor

Abstract

Soft robots and manipulators with inflatable links possess inherent safety characteristics that make them a very interesting choice for tasks requiring human-robot interaction. However, they may face important challenges in their performance due to low structural stiffness. Accurate positioning of the end effector may prove a difficult task due to limitations in gravity compensation, or due to the occurrence of oscillations during motion, especially in the absence of precise information about the payload that is being manipulated. In this paper a novel approach to the design and control of a robot with an inflatable link is proposed, where the link assumes the triple role of compliant structural element, touch sensor, and active mechanism to adjust the performance of vibration control algorithms. For this, a vision sensor is placed inside the link to provide information about the structural deformation. The sensor is used to measure the tip displacement, and also to detect contact of the link surface with the surrounding environment. The concept was implemented on a two-joint rigid manipulator with a single inflatable link. Experiments on vibration control and contact detection of the inflatable link are reported where the control system was able to significantly reduce tip oscillations, including when a payload was added to the tip. The robot was also able to make binary detection of contact between the inflatable link and the user, and change its operation mode accordingly.

Introduction

Robots with non-rigid structures offer important advantages in a number of tasks, ranging from search and rescue missions in catastrophe scenarios to the interaction with fragile living tissue in carefully controlled laboratory environments. Researchers working in the field of soft robotics have been trying to explore this concept in robots with a wide diversity of shapes, sizes and applications, that often draw their inspiration from biological systems [1], [2]. As robotics is facing an increasing demand for solutions that require human–robot interaction — to perform collaborative tasks in industry and also for assistive tasks in the health–care and home environments — a concept that has been gaining some attention is the use of inflatable structures in robotic manipulators. Inflatable manipulators can be designed to be extremely lightweight and also physically compliant. Both these characteristics contribute to safety when the tasks require proximity and interaction with people or with any fragile structure in general. Other advantages include: the materials and construction techniques tend to be relatively inexpensive when compared with other lightweight solutions; the device can be stored in a confined volume and deployed on site for usage; an increased payload–to–weight ratio can be expected from the lightweight construction. However, the major challenge posed by this type robot is control. The non-rigid nature of the structure makes it prone to oscillations during motion, or in case of accidental contact with the environment. The precision of end–effector positioning in the workspace is also reduced by the effect of gravity on the payload in static conditions. Therefore, in order to exploit the potential advantages of these devices, the control system will be required to compensate for all these issues.

Control of flexible manipulators, including flexible link and flexible joint robots, has been an important research topic for the past decades and there are several review papers available on the subject [3], [4], [5], [6]. This research has been mainly focused on manipulators with slender links made of common structural materials, like aluminum or fiber composite materials, with conventional electromagnetic actuators, although there are also examples of robots with links incorporating adaptive materials that perform structural control [7], [8].

Some of the first references to the utilization of inflatable links for robotic applications are [9], [10], [11], [12]. More recently Sanan et al. applied the concept to human–robot interaction. In [13] a basic design is presented, with a conventional single rotary joint, and a single inflatable link. A force sensor placed at the tip allowed a force control strategy to be implemented to evaluate the safety aspects of operation. The authors demonstrate stability of control when using a PD (Proportional Derivative) controller for the joint position. While feasibility of force control is also demonstrated, the point of contact is limited to correspond to the location of the force sensor, and the authors suggest the possibility of using printed strain gauges on the membrane of the link to extend the sensing capability to all its surface. In [14] a design with two inflatable links and three joints is presented. The first two joints are conventional rotary joints with DC motors at the base, and the third joint is actually a restriction applied to the tubular cross-section of the link actuated by tendons. This restriction reduces the local bending stiffness and divides the inflatable structure into three chambers, with the center chamber acting as the joint and both extremity chambers acting as links. The robot is capable of boolean contact detection with the second link, by comparing the measured pressure at the restricted section, used as a joint, with an expected pressure value for the given joint angle. This method implies that, while it is possible to detect contact with any point on the surface of the second link, the location of the point of contact remains unknown, thus limiting the range and efficacy of possible contact reaction schemes. Voisembert et al. developed a different kind of soft joint based on folding the inflatable link material, and proposed a robotic arm made from a single inflatable chamber. The structure is composed of straight segments acting as links, and joint segments that act as revolute joints by bending into the shape of a torus. Joint actuation is performed by tendons. In [15] focus was placed on the mechanical characterization of the novel joint segment, and it is shown that it has similar mechanical properties to the link segments in directions other than its axis. In [16] a finite-element study is presented showing that the volume of the inflatable chamber is kept nearly constant as a function of the joint angle. A numerical comparison of mechanical performance of this joint with that of [14] is also presented. The simulation showed significant influence of friction in the joint functioning. In [17] the detailed design of a long-range arm is addressed. Two different techniques to achieve the desired toroidal shape of the joint in bending are compared, one based on tensioned wire and the other based on stop-pins, the former achieving a better pitch angle. The existence and the nature of hysteresis in the actuation of the joints is preliminarily investigated. Qi et al. presented an inflatable arm with two joints and two links for a telepresence robot [18], [19]. The arm is composed of a single chamber with two section restrictions acting as the shoulder and elbow joints. Actuation is performed by three tendons per joint, allowing the control of both the joint angle and the direction of joint rotation. The kinematic modelling of the robot with soft inflatable joints is deduced and presented. However, the links are considered as rigid sections, without static or dynamic tip displacements between joints. The positioning accuracy was preliminarily investigated, and values close to 30mm in the X, Y and Z axes were obtained for an arm with a maximum extended length of 380mm (190mm each link). Kim et al. presented an inflatable robotic arm with four revolute joints and two links. The links are two separate inflatable cameras, and all the joints are made of rigid material with pneumatic actuation by bag actuators. In [20] the basic structure of the robot is presented, and joystic control is demonstrated. Each joint as a potentiometer for position feedback, but no measurement of the end-effector position is available. Point-to-point and trajectory tracking control of the robot are investigated in joint space. In [21] the authors address the problem of end-effector positioning by using visual feedback in eye-to-hand configuration. Only the last two joints of the robot are used, thus constraining the robot to move in a single plane. The performance of three different controllers is investigated under no-payload conditions, and an experiment with payload is also presented for the best controller. In both cases maximum positioning errors of less than 1mm were achieved for the best controller, but settling times close to 15s were observed. Killpack et al. have presented a pneumatically actuated, fully inflatable, soft humanoid robot. The pneumatic actuators consist of two bladders working in antagonistic form to produce the rotation of each joint. In [22] the authors address the control of a single joint using Linear Quadratic Regulator (LQR) and Model Predictive Control (MPC) techniques. These are then applied to the 5 joint arm of the humanoid robot. In the single joint set-up the angle is estimated from an inertial measurement unit, while in the complete arm the joint angles are calculated from the information of a motion capture system using light reflectors on the arm structure. In [23] the control of the single joint platform is developed further, and stiffness control for the joint is also introduced. The results show that the stiffness can be made to follow a reference signal with variation of up to 200%, while keeping the joint angle within a maximum error of 1deg. Results of varying stiffness during joint motion are also presented. In [24] several techniques that aim to improve the accuracy of position control of large-scale soft robots are presented, namely: a novel technique to estimate the homogeneous transformation between two different sensing systems, a method for rotational joint configuration estimation and a method for kinematic calibration. Experimental results of applying visual servoing techniques in a vicinity of the desired end effector position are also reported. These techniques allowed a reduction of Cartesian error between desired and measured end effector position from 6.7 cm to 2.2 mm. Overshoots between 2.5 % and 19.4% are reported, together with settling times between 4.9 s and 10.71 s. Results of the visual servoing tests in task space showed some signs of oscillations that, however, were not object of further study or interpretation.

The above mentioned literature was deliberately focused on robotic manipulators that use inflatable structures as passive elements in a serial kinematic chain. Outside this scope it is also possible to find other related designs like ”continuum” robot manipulators (e.g., [25], [26], [27]), where inflatable actuators responsible for the structure motion are almost indistinguishable from the soft structure itself, allowing the robot to extend or bend in a continuous shape. Although the dynamic issues posed by flexibility, and having pressure as a manipulated variable, are common to both types of robots, it is the authors view that the absence of passive links and the multidimensional nature of ”joint” displacements make these into a distinct research problem. It is nevertheless worth noting that some techniques developed for vibration control of ”classical” structures (in the sense that they are mostly built of passive elements) have been applied with some degree of success to continuum robots, like for instance in [28].

In previous works the passive inflatable structure is often chosen over rigid links because of its compliance and inherent safety. However, none of the above references within this scope seems to address the consequences of this compliance for the motion control of the structure itself, i.e., the control of the possible oscillations. Also, in spite of the safety of interaction with the complete surface of the inflatable link, contact with the structure was only addressed at either specific [13] or undetermined [14] locations. To extend this capability the use of printed strain gauges on the membrane has been previously suggested.

This paper addresses some of the challenges and advantages of using an inflatable link as part of the structure of a conventional serial chain manipulator. In particular the authors think their concept to be novel in that it assigns to the inflatable link the triple role of compliant structural element, touch sensor, and also of active component in tuning the performance of vibration control algorithms. This represents a significant structural optimization, that would be in contrast with a more straightforward type of solution consisting of separate components for linkage, touch sensing, and compensation mechanisms for the controllers. The concepts here introduced can be applied towards addressing this same challenges in a fully inflatable serial chain manipulator, or to the development of soft inflatable links specifically designed to be assembled, or added, as the last link of common rigid manipulators ensuring passive compliance of the end effector. Two important characteristics of inflatable structures are exploited for their advantages over other types of flexible manipulator links: 1) the soft hollow structure of the link is used as an encapsulation of a vision sensor, providing a structured environment for vision algorithms to detect tip displacement and also contact along the surface of the link; 2) The relation between the overall stiffness of the link and its internal pressure is used in a preliminary study addressing the performance of vibration control algorithms.

The main advantage of confining the vision sensor to a structured environment is to improve the image processing algorithms speed performance. Also, the placement of the vision sensor inside the inflatable structure precludes occlusion problems during interaction with the robot. These are pointed out in [5] as two of the main problems of using vision systems for control of flexible link manipulators. Placing the vision sensor inside the link is also a relatively simple and inexpensive solution to detect contact of the link with the surrounding environment. The complete surface of the link can potentially be used as a tactile sensor, without the added cost that tactile sensor networks or touch sensitive materials would have, as long as the contact forces generate a measurable deformation, and the contact surface lies within the field-of-view of the camera. Touch sensors that rely on the optical measurement of membrane deformations have also been proposed in literature in the past (e.g., in [29], [30], [31]) however, as they were not developed as structural elements, dynamics and potential use for dynamic control seems never to have been addressed by the authors. A recent review on touch sensing technologies, with a focus on vision-based tactile sensors, can be found in [32]. Finally, the fact that the structure is inflatable and its stiffness depends on internal pressure can also help the robot designer to address an important challenge posed by flexible link robots: that is to maintain accuracy and vibration control performance for different payloads or boundary conditions.