ICRA 2015 论文列表

The octopus uses the arm-swimming behavior primarily for escape, defense, or foraging. This mode of locomotion is comprised of two strokes, with the arms opening slowly and closing rapidly, and generally results in considerable propulsive acceleration. In light of the recent development by our group of an octopus-like eight-arm underwater robot, we are interested to analyze the details of the biological arm swimming motion, in order to understand its kinematics. In this paper, we address methodological aspects of the 3D reconstruction process of octopus arm trajectories, based on computer vision, and present the resulting arm swimming movement of a benthic common octopus. The 3D trajectories of all eight octopus arms were tracked and analyzed, providing information about speed, acceleration and arm elongation. The animal's performance is then used for a direct comparison with our 8-arm robotic swimmer. The data obtained provide new kinematic information about this, relatively unknown, propulsive mode, which can be exploited for multi-functional underwater robots.

This paper proposes a method for controlling a team of quadrotor micro aerial vehicles to perform agile maneuvers while holding a fixed relative formation, as well as transitioning between a sequence of formations. The objective is to coordinate the quadrotors to fly in intricate interlaced patterns, similarly to an air show demonstration team. The paper proposes a new abstraction, called a Virtual Rigid Body, which allows the quadrotors to hold relative positions while executing agile maneuvers as a group. By planning trajectories for the Virtual Rigid Body in SE(3), trajectories for each quadrotor are obtained in order to maintain the desired formation during the maneuver. The paper also proposes a method for sequencing a series of Virtual Rigid Body formations, and automatically designing collision free transitions between successive formations, while the team simultaneously executes a trajectory in SE(3). The resulting sequence of formations and transitions gives trajectories that weave intricate designs while avoiding collisions. The method is demonstrated experimentally with three KMel K500 quadrotors flying in a motion capture environment.

We consider the problem of camera pose estimation from a scene model obtained beforehand by a structure-from-motion (SfM) algorithm. The model is made of 3D points, each one of them being represented by its coordinates and a set of photometric descriptors such as SIFT, extracted from some of the input images of the SfM stage. Pose estimation is based on the matching of interest points from a test view with model points, using the descriptors. Descriptors having a limited invariance with respect to viewpoint changes, such an approach is likely to fail when the test view is far away from the images used to construct the model. Viewpoint simulation techniques, as ASIFT, have proved effective for wide-baseline image matching. This paper explores how these techniques can enrich a scene model by adding descriptors from simulated views, using either orthographic or pinhole virtual cameras. Viewpoint simulation permits pose estimation in situations where the approach based on the sole SIFT descriptors simply fails.

This paper presents a design, a control scheme, and experiments for a new vibrating magnet robot (ViMbot). Permanent magnets in the new robot vibrate due to an external vibrating magnetic field. The vibration motion of the robot induces the locomotion of the robot. The vibratory dynamic response of the ViMbot is heavily dependent on the frequency of the vibration. This paper shows that the direction of the robot locomotion can be controlled by the vibration frequency of the external magnetic field. To achieve the distinct dynamic responses depending on the vibration frequency, three acrylic plates with different torsional stiffness are attached at three sides of the triangle-shaped ViMbot and then a magnet is attached to each plate. A strategy to find the special frequencies inducing locomotion and associated experiments with the ViMbot are presented. Control experiments for the ViMbot using the special frequencies are also presented. Because the proposed robot is actuated by an external energy source, the robot can move without traditional actuation components such as motors with batteries. This new design concept allows us to miniaturize the proposed robot.

The ability to carry large payloads could greatly increase the applications of small, low cost climbing robots. We present a linear inchworm gait that uses a single powerful actuator to climb. To make this gait possible, we leveraged two new methods of achieving controllable, anisotropic adhesion (one method produces over 200 times stronger adhesion in the preferred direction). With controllable, anisotropic adhesion, the gait is robust to missed steps. In addition, the gait provides a stance in which the robot can rest without requiring power. An autonomous 9 gram robot is able to climb a smooth vertical surface at 3 mm/s, while hoisting more than a kilogram. We also present a scaled down version of the robot, which is considerably smaller than any previous dry adhesive climbing mechanism. It is actuated by externally powered Shape Memory Alloy, weighs 20 mg, and is capable of hoisting 500 mg. These climbers show that a large hoisting ability while climbing can be achieved using dry adhesives, and the presented concepts could aid in the development of autonomous, highly functional, small robots.

This paper describes a method of vehicle state prediction for an autonomous high-speed off-road Unmanned Ground Vehicle (UGV). Effective vehicle state prediction will allow a UGV to plan its navigation such that states (such as vertical acceleration induced by the terrain roughness) never exceed a desired threshold. In this paper a model of an n-wheeled generic vehicle is used determine its dynamics. Using a known terrain input profile the vehicle's output states are predicted using the developed n-wheel model. Simulated results of this vehicle state prediction approach are presented, as well as experimental tests using a UGV platform called Loc8. The experimental results used a 3D point cloud to determine the terrain input profile. Methods from the literature are tested against the developed n-wheeled vehicle state prediction method. Results show this n-wheel technique presents both advantages and disadvantages in comparison with existing techniques. The proposed approach predicts the average absolute acceleration much closer to the measured average absolute acceleration than existing approaches.

This paper introduces a bounding gait control algorithm that allows a variable-speed running in the MIT Cheetah 2. A simple impulse planning algorithm is proposed to design vertical and horizontal force profiles which make net impulse on the system during one cycle zero. This design of force profiles leads to the conservation of linear momentum over a complete step, providing periodicity in horizontal and vertical velocity. When designed profiles are applied to the system, periodic orbits with an ability to change running speed are obtained. A virtual compliance control in the horizontal and vertical direction has been added onto the designed force profiles to stabilize the periodic orbits. The experimental results show that the algorithm successfully achieved untethered 3D running of the MIT Cheetah 2, with speeds ranging from 0 m/sec to 4.5 m/sec on treadmills as well as on grassy fields.

Upper-body disabled people can benefit from the use of robot-arms to perform every day tasks. However, the adoption of this kind of technology has been limited by the complexity of robot manipulation tasks and the difficulty in controlling a multiple-DOF arm using a joystick or a similar device. Motivated by this need, we present an assistive vision-based interface for robot manipulation. Our proposal is to replace the direct joystick motor control interface present in a commercial wheelchair mounted assistive robotic manipulator with a human-robot interface based on visual selection. The scene in front of the robot is shown on a screen, and the user can then select an object with our novel grasping interface. We develop computer vision and motion control methods that drive the robot to that object. Our aim is not to replace user control, but instead augment user capabilities through our system with different levels of semi-autonomy, while leaving the user with a sense that he/she is in control of the task. Two disabled pilot users, were involved at different stages of our research. The first pilot user during the interface design along with rehab experts. The second performed user studies along with an 8 subject control group to evaluate our interface. Our system reduces robot instruction from a 6-DOF task in continuous space to either a 2-DOF pointing task or a discrete selection task among objects detected by computer vision.

This work presents a method for increasing the accuracy of standard visual inertial odometry (VIO) by effectively removing the angular drift that naturally occurs in feature-based VIO. In order to eliminate such drift, we propose to leverage the predominance of parallel lines in man-made environments by using the intersection of their image projections, known as vanishing points (VPs). First, an efficient inertial-based method is presented that accurately and efficiently detects such points. Second, a strategy to deal with these measurements within the framework of an EKF-based VIO system is presented. Furthermore, special care is taken in order to ensure the real-time execution of the estimator in order to comply with time-critical applications running on computationally constrained platforms. Experiments are performed in a mobile device on challenging environments and evaluated against the same VIO system without the use of VPs, demonstrating the superior accuracy when employing the proposed framework.

Detecting and interpreting contacts is a crucial aspect of physical Human-Robot Interaction. In order to discriminate between intended and unintended contact types, we derive a set of linear and non-linear features based on physical contact model insights and from observing real impact data that may even rely on proprioceptive sensation only. We implement a classification system with a standard non-linear Support Vector Machine and show empirically both in simulations and on a real robot the high accuracy in off- as well as on-line settings of the system. We argue that these successful results are based on our feature design derived from first principles.

The paper deals with the problem of planning movements of dual-arm anthropomorphic systems, with the aim of reducing the computational cost of the problem and making the movements look as human-like as possible. The key idea of the proposal is the search of synergies of the dual-arm anthropomorphic system in order to use them to reduce the dimension of the search space while preserving human-like appearance. This idea was already developed and successfully used to plan movements of robotic hands, thus the extension to a dual-arm system is attractive. The paper presents a description of the proposed approach as well as real experimental results that encourage doing further research in this line.

In this paper we introduce STEPPR (Sandia Transmission-Efficient Prototype Promoting Research), a bipedal robot designed to explore efficient bipedal walking. The initial iteration of this robot achieves efficient motions through powerful electromagnetic actuators and highly back-drivable synthetic rope transmissions. We show how the addition of parallel elastic elements at select joints is predicted to provide substantial energetic benefits: reducing cost of transport by 30 to 50 percent. Two joints in particular, hip roll and ankle pitch, reduce dissipated power over three very different gait types: human walking, human-like robot walking, and crouched robot walking. Joint springs based on this analysis are tested and validated experimentally. Finally, this paper concludes with the design of two unique parallel spring mechanisms to be added to the current STEPPR robot in order to provide improved locomotive efficiency.

The correct way to design controllers for dynamic robots is still very much an open question. This is in a large part due to the complexity and uncertainty in modeling their nonlinear dynamics. In this work, we focus on deriving concise dynamic expressions for a particular class of robots that can be used to better reduce uncertainty with respect to unknown parameters in realtime. We accomplish this by using an extended Kalman filtering framework in conjunction with an online controller that continuously maximizes a local measure of nonlinear observability. The main novel contribution of this work is that we directly use the nonlinear observability rank condition to derive the measure of observability at each time step. We are able to make this extension in part by focusing on serial-chain systems and exploiting the geometric structure in their dynamic models. In particular, we derive concise, closed-form and exact analytical representations for the forward dynamics, linearization, and nonlinear observability rank condition of a fixed-base serial manipulator with actively controlled elastic joints. An example is presented in which the spring constants and damping coefficients for a series-elastic actuated manipulator are estimated using the online observability maximizing techniques we derive.

Socially interacting robots will need to understand the intentions and recognize the behaviors of people they come in contact with. In this paper we look at how a robot can learn to recognize and predict people's intended path based on its own observations of people over time. Our approach uses people tracking on the robot from either RGBD cameras or LIDAR. The tracks are separated into homogeneous motion classes using a pre-trained SVM. Then the individual classes are clustered and prototypes are extracted from each cluster. These are then used to predict a person's future motion based on matching to a partial prototype and using the rest of the prototype as the predicted motion. Results from experiments in a kitchen environment in our lab demonstrate the capabilities of the proposed method.

Scene modeling is an important first stage in visual robot localization. In recent years, the bag-of-words (BoW) scene modeling approach has attracted considerable attention as a method for obtaining compact discriminative scene descriptors for map retrieval. However, a BoW scene descriptor alone cannot address partial view changes and often produces poor localization in practice. In this work, we address this issue by proposing a simple effective approach, “unsupervised part-based scene modeling,” in which a set of useful parts is discovered via scene parsing and the parts are used as additional queries for the map retrieval. We also address the issue of discovering useful parts in a scene, and present a solution that provides similar parts for similar scenes. The next contribution of this work is that we present a practical robot self-localization system that consists of three distinct steps: (1) robust hierarchical scene parsing to obtain multiple scene/part queries, (2) saliency-based selection of useful parts, and (3) aggregation of ranking results from multiple scene/part queries to obtain a reliable ranking result. For rank aggregation, we consider multiple search engines for multiple part queries and adopt the idea of unsupervised rank fusion. Experimental results obtained using a challenging outdoor scene dataset show that our approach is an improvement over previous approaches despite the fact that we do not rely on domain-specific scene/part models nor supervision.

In order for robots to be useful in real world learning scenarios, non-expert human teachers must be able to interact with and teach robots in an intuitive manner. One essential robot capability is wide-area (mobile or nonstationary) pick-and-place tasks. Even in its simplest form, pick-and-place is a hard problem due to uncertainty arising from noisy input demonstrations and non-deterministic real world environments. This work introduces a novel method for goal-based learning from demonstration where we learn over a large corpus of human demonstrated ground truths of placement locations in an unsupervised manner via Gaussian Mixture Models. The goal is to provide a multi-hypothesis solution for a given task description which can later be utilized in the execution of the task itself. In addition to learning the actual arrangements of the items in question, we also autonomously extract which frames of reference are important in each demonstration. We further verify these findings in a subsequent evaluation and execution via a mobile manipulator.

In this paper we propose a novel hybrid Cartesian force/impedance controller that is equipped with energy tanks to preserve passivity. Our approach overcomes the problems of (hybrid) force control, impedance control, and set-point based indirect force control. It allows accurate force tracking, full compliant impedance behavior, and safe contact resemblance simultaneously by introducing a controller shaping function that robustly handles unexpected contact loss and avoids chattering behavior that switching based approaches suffer from. Furthermore, we propose a constructive way of initiating the energy tanks via the concept of task energy. This represents an estimate of the energy consumption of a given force control task prior to execution. The controller can be applied to both rigid body and flexible joint dynamics. To show the validity of our approach, several simulations and experiments with the KUKA/DLR LWR-III are carried out.

In recent years impressive results have been presented illustrating the potential of quadrotors to solve challenging tasks. Generally, the derivation of the controllers involve complex analytical manipulation of the dynamics and are very specific to the task at hand. In addition, most approaches construct a trajectory and then design a stabilizing controller in a separate step, whereas a fully optimal solution requires finding both simultaneously. In this paper, a generalized approach is presented using an iterative optimal control algorithm. A series of complex tasks are thus solved using the same algorithm without the need for manual manipulation of the system dynamics, heuristic simplifications, or manual trajectory generation. First, aggressive maneuvers are performed by requiring the quadrotor to pass with a slung load through a window not high enough for the load to pass while hanging straight down. Second, go-to-goal tasks with single and double rotor failure are demonstrated. The adaptability and applicability of this unified approach to such diverse tasks with a nonlinear, underactuated, constrained, and in the case of the slung load, hybrid quadrotor systems is thus shown.

Autonomous manipulation of objects requires reliable information on robot-object contact state. Underwater environments can adversely affect sensing modalities such as vision, making them unreliable. In this paper we investigate underwater robot-object contact perception between an autonomous underwater vehicle and a T-bar valve using a force/torque sensor and the robot's proprioceptive information. We present an approach in which machine learning is used to learn a classifier for different contact states, namely, a contact aligned with the central axis of the valve, an edge contact and no contact. To distinguish between different contact states, the robot performs an exploratory behavior that produces distinct patterns in the force/torque sensor. The sensor output forms a multidimensional time-series. A probabilistic clustering algorithm is used to analyze the time-series. The algorithm dissects the multidimensional time-series into clusters, producing a one-dimensional sequence of symbols. The symbols are used to train a hidden Markov model, which is subsequently used to predict novel contact conditions. We show that the learned classifier can successfully distinguish the three contact states with an accuracy of 72% ± 12 %.