IROS 2015 Papers

A novel fluidic actuation system has been developed for in situ imaging of anatomic tissues. The actuator consists of a micromachined superelastic tool guide driven by a pair of pneumatic artificial muscles. Two additional working channels allow easy interchange of instruments or sensing equipment. This paper describes the design and construction of the actuation system. Experimental results are also reported indicating a bending repeatability of 0.1 degrees and an operational bandwidth exceeding 8Hz. To show-case the performance of the device, the actuator was loaded with an all-optical ultrasound imaging probe. First scanned images of human placental tissue surface using an all-optical ultrasound probe are presented. While a model has been developed to estimate the probe position in space as function of the input pressure, in future work, this model will be complemented with additional sensor measurements of the bending probe taking into account the hysteretic behaviour of both muscles and nitinol structure.

Teaching robots manipulative skills through human demonstration is an important research problem and can be used to quickly program robots in future manufacturing industries. To understand human demonstration, manipulative actions need to be recognized. To improve the recognition performance, we use three kinds of sensors to capture the motion and force involved in the fine manipulative actions. In addition, by taking advantage of the action/object correlation, the recognition accuracy can be further improved. In the proposed approach, important features for individual actions are selected first. Hidden Markov Models (HMMs) are employed to characterize the temporal changes. Then, a Bayesian model is adopted to model the object/action dependency. Our approach was evaluated through experiments on assembly tasks. The experimental results show that the proposed approach can recognize manipulative actions effectively.

Autonomous systems rely on an increasing number of input sensors of various modalities, and the problem of sensor fusion has received attention for many years. Autonomous system architectures are becoming more complex with time, and the number and placement of sensors will be modified regularly, sensors will fail for many reasons, information will arrive asynchronously, and the system will need to adjust to rapidly changing environments. To address these issues we propose a new paradigm for fusing information from multiple sources that draws from the rich of field pertaining to financial markets, particularly recent research on prediction market design. Among the many benefits of this financialized approach is that, both in theory and in practice, markets are well-equipped to robustly synthesize information from diverse sources in a decentralized fashion. Our framework poses sensor processing algorithms as profit-seeking market participants, data is incorporated via financial transactions, and the joint estimation is represented as a price equilibrium. We use pedestrian detection as a motivating application. Pedestrian detection is a well studied field and essential to autonomous driving. Real world fusion results are presented on RGB and LIDAR data from the KITTI Vision Benchmark Suite. We demonstrate we can achieve comparable performance to state-of-the-art hand designed fusion techniques using the proposed approach.

We developed an information-providing robot for a shopping mall. It moves around and hands over flyers about the mall and when people are interested or approach the robot to obtain some information, it starts the information-providing service. During the information-providing service, it provides directions and recommendations of shops in the mall. The robot is developed to be fully autonomous, but due to low performance in automatic speech recognition, this function is overtaken by a human operator. In our field trial, the robot worked for 53.4 hours (over 11 days) without any major failure (83.5% of time autonomously operated). We conducted interviews with visitors, and revealed that the majority of them found the information from the robot useful, and wanted to use it again.

We present the design, fabrication, and characterization of a fiber optically sensorized robotic hand for multi purpose manipulation tasks. The robotic hand has three fingers that enable both pinch and power grips. The main bone structure was made of a rigid plastic material and covered by soft skin. Both bone and skin contain embedded fiber optics for force and tactile sensing, respectively. Eight fiber optic strain sensors were used for rigid bone force sensing, and six fiber optic strain sensors were used for soft skin tactile sensing. For characterization, different loads were applied in two orthogonal axes at the fingertip and the sensor signals were measured from the bone structure. The skin was also characterized by applying a light load on different places for contact localization. The actuation of the hand was achieved by a tendon-driven under-actuated system. Gripping motions are implemented using an active tendon located on the volar side of each finger and connected to a motor. Opening motions of the hand were enabled by passive elastic tendons located on the dorsal side of each finger.

Friction along the Bowden-cable transmission degenerates control performance unless it is properly compensated. Friction is produced when the bending angle of the Bowden-cable changes as the relative position of the actuator and the end-effector changes. This study proposes a method, termed loop routing, to compensate friction along the Bowden-cable. Loop routing involves making a one-round loop along the sheath that continuously maintains the sheath's bending angle at 2π regardless of the end-effector's position in 2-D space. This minimizes the bending angle change of the sheath as the end-effector translates in a 3-D workspace, which minimizes the friction change and enables feedforward friction compensation of the Bowden-cable without employing a sensor. An experiment in open-loop tension control of the Bowden-cable was conducted to evaluate the performance of the proposed method. Results show that the output tension follows the reference tension well, with an RMS error of 4.3% and a peak error of 13.3% of maximum reference.

Here we present a non-holonomic distance function for unicycle-type vehicles, and use this distance function to extend the optimal path planner RRT* to handle nonholonomic constraints. The critical feature of our proposed distance function is that it is also a control-Lyapunov function. We show that this allows us to construct feedback policies that stabilizes the system to a target pose, and to generate the optimal path that respects the non-holonomic constraints of the system via the non-holonomic RRT*. The composition of the Lyapunov function that is obtained as a result of this planning process provides stabilizing feedback and the cost-to-go to the final destination in the neighborhood of the planned path, adding much flexibility and robustness to the plan.

Full autonomy remains a challenge for miniature robotic platforms due to mass and size requirements of on-board power and control electronics. This paper presents a solution to these challenges with a 2.3g autonomous legged robot. An off-the-shelf optical mouse sensor is adapted for use on the Harvard Ambulatory Microrobot (HAMR) by reducing the sensor weight by 36% and achieving a position error below 11% when suspended 3mm above a cardstock surface. The position data is combined with data from a gyroscope for feedback control of both position and orientation. A microcontroller processes the sensor data and commands a controlled gait to HAMR that is powered by a battery, a boost converter and high voltage drive electronics. Solar cells are used as an alternative source providing enough power for autonomous operation of the robot. The resulting deviation for a controlled straight-line walk using both sensors to minimize lateral deviation and angular error is only 4.6%, compared to an error of 31% in an uncontrolled, straight-line walk.

Adaptation is an essential capability for intelligent robots to work in new environments. In the learning framework of Programming by Demonstration (PbD) and Reinforcement Learning (RL), a robot usually learns skills from a latent feature space obtained by dimension reduction techniques. Because the latent space is optimized for a specific environment during the training phase, it typically contains fewer variations. Accordingly, searching for a solution within the latent space can be less effective for robot adaptation to new environments with unseen changes. In this paper, we propose a novel Feature Space Decomposition (FSD) approach to effectively address the robot adaptation problem, which is directly applicable to the learning framework based on PbD and RL. Our FSD method decomposes the high-dimensional original features extracted from the demonstration data into principal and non-principal feature space. Then, the non-principal features are used to form a new low-dimensional search space for autonomous robot adaptation based on RL, which is initialized using a generalized trajectory represented by a Gaussian Mixture Model that is learned from the principal features. The scalability of our FSD approach guarantees that optimal solutions can be found in the new non-principal space, if they exist in the original feature space. Experimental results on real robots validate that our FSD approach enables the robots to effectively adapt to new environments, and is usually able to find optimal solutions more quickly than traditional approaches when significant environment changes occur.

The absence of touch of sense is a widely known drawback of robotic minimally invasive surgery (MIS). This paper proposes a design of optic soft tactile arrays which is promising to be adapted for MIS. The proposed design consists of multiple soft material channels. Each channel is designed using the Bernoulli pipe structure to amplify the sensor's sensitivity through input and output diameter difference. A multi-core optic fiber cable and a camera are used to capture the change of light intensity caused by the contact forces applied onto the individual soft material channels. The proposed sensor has the following advantages: 1) making use of 3D printing and soft material casting, it is suitable for designing sensors with high density of tactile elements; 2) it also allows the sensor to be designed in an arbitrary shape to fit various MIS applications; 3) compared to other light-intensity based tactile sensor, it is easy to fabricate and miniaturize; it avoids the complexity of attaching reflectors to individual sensing elements; 4) it is immune to electromagnetic interference. In this paper, a prototype which has 3×3 tactile elements in an area of 9.5 × 11 mm2 has been developed and test for feasibility study. Also, a noise-filtering algorithm is developed to reduce the imaging noise. Validation experiments were carried out and results show that the average measurable force range for a single tactile element is 0 to 1.622N with an average accuracy of 97%. The sensor has low crosstalk-to-signal ratio, 1.8% on average, and has no signal drift over time.

This paper presents a novel indicator for lump detection that uses only contact pressure distribution, as well as its feasibility evaluation. The conventional lump detection methods are based on a pressure map obtained from contact pressure distributions, but they requires an adequate regulation of the contact condition during measurements. An alternative approach is to use a stiffness map, for which displacement measurement is necessary using additional devices or precise motion control. The proposed indicator relies on only the difference between the normal force and the surface normal obtained from contact pressure distributions. Thus, contact condition or probe motion does not affect the performance of the proposed method. The feasibility of the proposed method was evaluated by simulations and experiments, both of which showed promising results.

This paper presents a fault tolerant control strategy based on an offline control mixing for an octorotor unmanned aerial vehicle (UAV) regarding several rotor failures. This strategy consists of a set of explicit laws, computed offline, each one dedicated to a fault situation. The corresponding law is selected according to the output of a fault detection and isolation (FDI) module. This module is designed with a non-linear sliding mode observer. The main advantage of this architecture is the deterministic character of the solution, its fast operation and the low computational load. The effectiveness of this approach is illustrated through real experimental application to a coaxial octorotor, where up to four motor failures are considered.

Maintaining clearance, or distance from obstacles and sampling efficient enough configurations on the medial axises are a vital component for successful motion planning. Maintaining high clearance often creates safer paths for robots. Having bias for sampling on medial axis also offers higher possibility to find a path in complex environment where the feasible configuration space only occupies a small proportion of the whole space. Inspired by the similarity between medial axis and max-margin scheme in optimization, especially in Support Vector Machine, we propose a new method to quickly construct the medial axis for the motion planning environment both in low and high dimensional space. However, directly applying the SVM classification on the large volume of uniformly sampled configurations suffers from huge computation and the medial axis is usually not the real medial axis due to SVM's optimization function's tolerance to the mis-classification. Instead, we show a method that can quickly push any configuration to the medial axis by using the characteristics of the Max-Margin's optimization function. Experiments in low and high dimensional space and comparisons with other medial-axis motion planning algorithm are shown.

New sensor systems, efficient planning algorithms and increased computational power have led to a growing interest in high-resolution 3-D data for challenging applications like mobile manipulation, 3-D mapping and object recognition. 3-D laser scanners built using a rotating or swivelling 2-D laser scanner are a widespread technology for obtaining 3-D point clouds, but convenient calibration methods are needed to increase the sensor precision and loosen the requirements on mechanical tolerances. We present an approach for the automatic self-calibration of such scanners, using only a single scan of a targetless environment recorded over a complete 360° rotation of the external motor that controls the motion of the 2-D sensor. By exploiting the intrinsic redundancies of the recorded point clouds and decimating the clouds using a voxel grid filter, we are able to compute and optimize a point cloud quality measure very quickly and without relying on explicit calibration targets. Our results on simulated and real data show the effectiveness of our approach by creating high-quality 3-D point clouds.

Accurate surface normal computation is one of the most basic and important tasks for 3d perception. While much progress has been made in speeding up normal estimation algorithms and improving their accuracy, a significant inaccuracy still remains even with modern implementations, which is the correct determination of surface normals close to non-differentiable surface edges. Current algorithms tend to amalgamate neighborhood points from independent surfaces yielding normals that neither fit well to the one nor the other surface. This paper introduces a fast and accurate 3d edge detection algorithm suitable to detect discontinuities both in depth and on surfaces with nearly 90% accuracy at rates beyond 30 Hz. Based on this method, we demonstrate how established normal estimation algorithms can be extended for edge-awareness. Additionally, a new edge-aware, fast, accurate, and robust normal estimation approach is described which exploits the data structures computed for 3d edge detection and estimates normals at 23 Hz. We assess the performance of all proposed methods and compare them with other state-of-the-art approaches.

T3D edge detection from a depth image is an important technique of 3D object recognition in preprocessing. There are three types of 3D edges in a depth image called jump, convex roof, and concave roof edges. Conventional 3D edge detection based on ring operators has been proposed. The conventional ring operator can detect three types of 3D edges by classifying the response of Fourier transforms. Since the conventional method needs to apply Fourier transforms to all pixels of a depth image, real-time processing cannot be done due to high computational cost. Therefore, this paper presents a fast and reliable method of detecting three types of 3D edges by using a decision tree. The decision tree is trained under supervised learning from numerous synthesized depth images and labels by capturing depth relations between candidate pixels and pixels on a ring operator to classify 3D edges. The experimental results revealed that the proposed method has 25 times faster than the conventional method. This paper also presents some examples of 3D line and 3D convex corner detection based on results obtained with the proposed method.

Members of a team are able to coordinate their actions by anticipating the intentions of others. Achieving such implicit coordination between humans and robots requires humans to be able to quickly and robustly predict the robot's intentions, i.e. the robot should demonstrate a behavior that is legible. Whereas previous work has sought to explicitly optimize the legibility of behavior, we investigate legibility as a property that arises automatically from general requirements on the efficiency and robustness of joint human-robot task completion. We do so by optimizing fast and successful completion of joint human-robot tasks through policy improvement with stochastic optimization. Two experiments with human subjects show that robots are able to adapt their behavior so that humans become better at predicting the robot's intentions early on, which leads to faster and more robust overall task completion.

Spring loaded inverted pendulum (SLIP) model used simple spring mass mechanism to explain leg function and ground reaction force in legged locomotion. Balancing the upper body can be addressed by addition of a rigid trunk to this template model. The resulting model is not conservative and needs hip torque to keep the trunk upright during locomotion, like humans. Leg force modulated compliant hip (FMCH) is our new model for postural control in walking which employs the leg force feedback to adjust the hip compliance. Such an application of positive force feedback presents a new template for neuromuscular model. This method provides stable and robust walking in simulations and also mimics human-like kinetic behavior. Analyzing human walking experiment shows that FMCH can explain the hip torque-angle relation for different walking speeds. Finally, this approach may physically implement the virtual pendulum (VP) concept, observed in human/animal locomotion.

Teleoperated Unmanned Ground Vehicles (UGVs) are expected to play an important role in future search and rescue operations. In such tasks, two factors are crucial for a successful mission completion: operator situational awareness and robust network connectivity between operator and UGV. In this paper, we address both these factors by extending a new Free Look Control (FLC) operator interface with a graphical representation of the Radio Signal Strength (RSS) gradient at the UGV location. We also provide a new way of estimating this gradient using multiple receivers with directional antennas. The proposed approach allows the operator to stay focused on the video stream providing the crucial situational awareness, while controlling the UGV to complete the mission without moving into areas with dangerously low wireless connectivity. The approach is implemented on a KUKA youBot using commercial-off-the-shelf components. We provide experimental results showing how the proposed RSS gradient estimation method performs better than a difference approximation using omnidirectional antennas and verify that it is indeed useful for predicting the RSS development along a UGV trajectory. We also evaluate the proposed combined approach in terms of accuracy, precision, sensitivity and specificity.