IROS 2017 Papers

In the framework of Human-Robot Collaboration, a robot and a human operator may need to move in close coordination within the same workspace. A contactless coordinated motion can be achieved using vision, mounting a camera either on the robot end-effector or on the human. We consider here one instance of such a visual coordination task, with the robot end-effector that should maintain a prescribed position with respect to a moving RGB-D camera while pointing at it. For the 3D localization of the moving camera, we compare three different techniques and introduce some improvements to the best solution found for our application. For the motion tracking problem, we introduce a relaxed version of the pointing part of the task. This allows to take advantage of the redundancy of the robot, distributing the control effort over the available degrees of freedom. The effectiveness of the proposed approach is shown by V-REP simulations and experiments with the 7-dof KUKA LWR manipulator.

In this paper, we address the problem of using a camera with limited field of view for controlling the motion of a quadrotor in aggressive flight regimes. We present a minimum time trajectory planning method that guarantees visibility of the image features while allowing the robot to undertake aggressive motions for which the usual near-hovering assumption is violated. We exploit differential flatness and B-Splines to parametrize the system trajectories in terms of a finite number of control points, which can then be optimized by Sequential Quadratic Programming (SQP). The control strategy is similar to a Receding Horizon Control (RHC) approach for modifying online the reference trajectory in order to account for noise, disturbances and any non-modeled effect. The algorithm is validated in a physically realistic simulation environment.

We propose a reactive locomotion strategy, called height reflex, that is useful to address big elevation changes in the terrain (e.g. when a quadruped robot has to step down from a high platform). In these cases the swing leg can lose mobility creating issues in the subsequent steps. The height reflex is a foot trajectory replanning strategy that redistributes the swing motion (in a smart way) to the stance legs to “lower” the whole trunk and to aid the foothold searching motion. To spread the motion we exploit a massless link model of the robot with virtual dampers at the joints, which is used to replan the feet trajectories. The proposed approach is able to incorporate kinematic limits, it is easy-to-tune, computationally efficient and suitable for real-time implementations. The reflex is implemented and experimentally evaluated on the 80 kg hydraulic quadruped HyQ. With our approach we were able to address high steps, up to 24 cm which is 30% of HyQ leg length and 53% of its retractable leg range.

We present a learning-based mapless motion planner by taking the sparse 10-dimensional range findings and the target position with respect to the mobile robot coordinate frame as input and the continuous steering commands as output. Traditional motion planners for mobile ground robots with a laser range sensor mostly depend on the obstacle map of the navigation environment where both the highly precise laser sensor and the obstacle map building work of the environment are indispensable. We show that, through an asynchronous deep reinforcement learning method, a mapless motion planner can be trained end-to-end without any manually designed features and prior demonstrations. The trained planner can be directly applied in unseen virtual and real environments. The experiments show that the proposed mapless motion planner can navigate the nonholonomic mobile robot to the desired targets without colliding with any obstacles.

A truly autonomous mobile robot have to solve the SLAM problem (i.e. simultaneous map building and pose estimation) in order to navigate in an unknown environment. Unfortunately, a universal solution for the problem hasn't been proposed yet. The tinySLAM algorithm that has a compact and clear code was designed to solve SLAM in an indoor environment using a noisy laser scanner. This paper introduces the vinySLAM method that enhances tinySLAM with the Transferable Belief Model to improve its robustness and accuracy. Proposed enhancements affect scan matching and occupancy tracking keeping simplicity and clearness of the original code. The evaluation on publicly available datasets shows significant robustness and accuracy improvements.

The visual difference between outcomes in many robotics tasks is often subtle, such as the tip of a screw being near a hole versus in the hole. Furthermore, these small differences are often only observable from certain viewpoints or may even require information from multiple viewpoints to fully verify. We introduce and compare three approaches to selecting viewpoints for verifying successful execution of tasks: (1) a random forest-based method that discovers highly informative fine-grained visual features, (2) SVM models trained on features extracted from pre-trained convolutional neural networks, and (3) an active, hybrid approach that uses the above methods for two-stage multi-viewpoint classification. These approaches are experimentally validated on an IKEA furniture assembly task and a quadrotor surveillance domain.

In grasp detection, the robot estimates the position and orientation of potential grasp configurations directly from sensor data. This paper explores the relationship between viewpoint and grasp detection performance. Specifically, we consider the scenario where the approximate position and orientation of a desired grasp is known in advance and we want to select a viewpoint that will enable a grasp detection algorithm to localize it more precisely and with higher confidence. Our main findings are that the right viewpoint can dramatically increase the number of detected grasps and the classification accuracy of the top-n detections. We use this insight to create a viewpoint selection algorithm and compare it against a random viewpoint selection strategy and a strategy that views the desired grasp head-on. We find that the head-on strategy and our proposed viewpoint selection strategy can improve grasp success rates on a real robot by 8% and 4%, respectively. Moreover, we find that the combination of the two methods can improve grasp success rates by as much as 12%.

Recent progress in LOC technology have made it possible to analyze cells in microfluidic devices in a microscopic environment. Vision-based cell analysis systems have a limitation in their analysis range owing to the small view area of the microscopic objective lens; thus, an expansion of the view area of microscopic observation is required. In this paper, we propose a view expansion system for photomicroscopy using the observing point movement of a Galvano mirror. We develop a prototype of the proposed view expansion system that consists of a microscope, Galvano mirror, and vision system. Then, the visual field of the developed microscopic view expansion system is derived experimentally. Finally, our system is demonstrated through panoramic image developing experiments. The effectiveness of the proposed system is verified by comparing it with a normal microscopic video shooting area.

In this study, a variable viscoelastic joint system comprising antagonized artificial muscles and magneto-rheological fluid brakes is proposed. This system enables human assisting devices such as robotic exoskeletons to retain structural softness when compared with the existing devices driven by motors and reduction gears, which can only achieve superficial softness. The authors had proposed the system and showed its effectiveness via experiments using a prototype with one degree of freedom in their previous works. In this study, as the next step in designing the human assisting device, a lower body assistive prototype “Airsist” is introduced. In addition, a control method that can cooperate with the wearer without the need for an additional operation is proposed.

This paper introduces a new class of self-reconfigurable robot: the variable topology truss (VTT). Related to an existing class of robots, the variable geometry truss (VGT), variable topology trusses have the additional capability to change the topology of the truss through self-reconfiguration. The hardware necessary to achieve this is introduced, and the constraints and capabilities of this new type of robot are analyzed by introducing the concept of a topology neighbor graph. Lastly, the minimal reconfigurable VTTs, which require 18 members, are identified and their achievable topologies are enumerated.

Admittance control is a widely used approach for guaranteeing a compliant behavior of the robot in physical human-robot interaction. When an admittance-controlled robot is coupled with a human, the dynamics of the human can cause deviations from the desired behavior of the robot, mainly due to a stiffening of the human arm, and thus generate high-frequency unsafe oscillations of the robot. In this paper we present a novel methodology for detecting the rising oscillations in the human-robot interaction. Furthermore, we propose a passivity-preserving strategy to adapt the parameter of the admittance control in order to get rid of the high-frequency oscillations and, when possible, to restore the desired interaction model. A thorough experimental validation of the proposed strategy is performed on a group of 26 users performing a cooperative task.

Construction automation has historically been driven by top-down implementations of specific tasks, which are neither responsive nor resilient to dynamic situations, and often require centralized control or human supervision. Previous work on robotic assembly has generally neglected to consider forces acting on the structure, whether in the completed structure alone or throughout the building process. In this paper, we investigate the utility of local force measurements in guiding construction by a distributed team of strut-climbing robots, focusing on a scenario involving building an unsupported span out across a gap in a two-dimensional vertical plane, as a step towards building a bridge. We show that such measurements enable robots to build structures that cantilever significantly further than those built by robots without access to such information, while maintainig stability throughout the building sequence. We consider both structures securely anchored to the ground and those resting unanchored atop it, using a counterbalancing approach in the latter case to permit cantilevering. The principles explored in simulation are also demonstrated in hardware, including a prototype strut-climbing robot and truss components, incorporating a cost-effective sensor implementation that reports the requisite force information.

We propose a novel approach for safe tracking of a moving target in cluttered environments using a quadrotor. The key contribution of our work is a formulation that enables the generation of safe and dynamical feasible tracking trajectories that satisfy arbitrary relative motion patterns (circling, parallel tracking, undirectional tracking, etc.) with respect to the target. In our framework, forming the desired relative motion pattern between the quadrotor and the target only requires a generative function that specifies relative positions at different time instants. Our method generates samples to fit a piecewise-polynomial representation of the desired relative motion pattern and embeds it into a cost function for solving valid tracking trajectories via quadratic programming. Collision avoidance is achieved by squeezing the trajectory into a collision-free flight corridor, and dynamical feasibility is achieved by enforcing bounds on corresponding derivatives. Both of which can be written as linear constraints for the quadratic programming. Our approach is lightweight and can be implemented for real-time target tracking. We use a simulated cluttered environment and multiple desired relative motion patterns to demonstrate the performance of the proposed approach.

Pneumatic Artificial Muscle (PAM) actuators have been used as exoskeletons because of their inherited compliance and high power-weight ratio. However, creating accurate models remains difficult mainly due to the compliance issue; the model can be changed by the force applied by the user. Therefore, both user and robot actions need to be considered for sufficient excitation of PAMs that are equipped in exoskeleton robots, unlike typical rigid actuators that can only be sufficiently excited by robot actions. In this paper, we propose a user-robot collaborative excitation approach for PAM model identification as an active learning framework for sequentially collecting data by deriving and executing optimal user and robot actions at each step with Gaussian processes. The optimal actions, which are executed by the robot, are displayed on a monitor that enables the user to execute them. We conducted experiments using a powered elbow exoskeleton with a PAM actuator. Experimental results show that our method can more efficiently identify the PAM model than a standard model identification method that does not use any data acquired through user-robot collaboration.

Haptic feedback for minimally invasive surgery is a widely debated topic, especially force sensors for this kind of application are thoroughly researched. The current state-of-the-art includes several realization options, but none of them are suitable to be integrated in small diameter instruments. In this work, we present a new approach for the generation of haptic feedback by using two separate feedback mechanisms for spatial kinesthetic and unidirectional tactile feedback in a teleoperation system with a parallel kinematic slave robot and the design of the corresponding user interfaces. Kinesthetic forces are measured with proximal sensors in the robot kinematic and displayed with a delta-type kinematic mechanism. Tactile signals are acquired with an acceleration sensor placed closely at the tip of the robot manipulator and conveyed open-loop by a vibrating cylinder to the user's palm. Kinesthetic forces up to 15 N can be displayed up to a frequency of 50 Hz. The tactile interface displays accelerations up to 15m/s for frequencies from 50 Hz to 1000 Hz. The combination of the two interfaces allows the display of high-fidelity haptic feedback with greatly reduced requirements on the interface components.

The successes of teleoperation scenarios for mobile robots depends on a stable and reliable communication link. The environment information collected by the robot - represented by 2D or 3D images - has to be provided with a high resolution and a low delay to ensure a fast and precise system response. But in most realistic applications, the communication parameters fluctuate strongly over time. It is necessary to monitor the communication link continuously to react in case of reduced bandwidth and increased delay. But which environment information and correspondingly which bandwidth is necessary to control a robot safely? Due to a missing reliable rule set we investigated this question for a UAV scenario based on two different environment representations (camera images, gridmaps). We designed a simulator based study and evaluated the capability of the participants to control a robot in case of delayed or coarsely rasterized information. Although our study involved only non-experts, we found some interesting first results. While the performance of our participants depends strongly on the delay, it is nearly independent on the image resolution, which suggests downsampling as a valid response for bandwidth decrease. We also found that participants generally struggle with using grid map for controlling the robot. However, this type of interfaces requires far less bandwidth than images. They also excel in situations with higher delays, which makes them the tool of choice when there are really bad channel conditions.

This paper proposes a User Interface (UI) of a force-controlled arm for endoscopic surgery. It consists of two input configurations for endoscope type (straight or oblique endoscope) and operation type (direct or indirect operation). Direct operation is used for wide area motion based on direct operation of the force-controlled arm, and indirect operation is for fine and precise movement to adjust the visual field by tool tracking or joystick manipulation. These were tested by doctors and development members as subjects. It was then confirmed that both operation types can be effective. We also discuss about the importance of detecting contact with organs from outside the patient's body (without sensors near the tool tip) and evaluate the feasibility of this feature with external force estimation function for force-controlled arm as a first step. As a result, it was found that the external forces in the insertion direction to the trocar can be estimated with an error of about ±0.2 N and up to 1.0 N on the other directions. Therefore we confirmed that contact detection function can be effective.

Focusing on upper limb rehabilitation of weak stroke patients, this pilot study explores how motion intent can be detected using force sensitive resistors (FSR). This is part of a bigger project which will see the actuation and control of an intent-driven exoskeleton. The limited time stroke survivors have with their therapists means that they can not often get enough training. During active-assisted training, therapists guide the paralysed limb through a movement only after detecting visual or haptic cues of the motion intent from the patient. Aiming to replicate therapist practices of recognising patients' intention to move, a pilot study of a tactile system is performed. The system will perform consistently even with patients who have low muscle strength and control ability. Currently available devices for detecting muscle activity do not offer the robustness and performance necessary; Electromyography (EMG) sensors, a well-established method, is affected by factors like skin moisture and BCI (Brain Computer Interface) has a slow response time. The proposed tactile sensing system is a simple yet robust solution both from a sensing as well as a usability point of view. Pilot experiments have been performed with a healthy subject emulating low muscle activation conditions. An overall accuracy of 80.45% is achieved when detecting forearm and arm muscle contractions and hence motion intent.

In this paper, we first consider a simple Bayesian fusion problem in a matrix Lie group, and propose to tackle it using the unscented transform. The method is then leveraged to derive two simple alternative unscented Kalman filters on Lie groups, for both cases of noisy partial measurements of the state, and full state noisy measurements of the state on the group. The general method is applied to a robot localization problem, and results based on experimental data combined with extensive Monte-Carlo simulations at various noise levels illustrate the superiority of the approach over the standard UKF.