IROS 2015 Papers

Existing commercial driver assistance systems, including automatic braking systems and lane-keeping systems, may monitor the state of the vehicle or the environment to determine whether the systems should intervene. However, the state of the human driver is not typically included in the decision making process. In this paper, we propose to use hidden mode stochastic hybrid systems to model the interaction between the human driver and the vehicle. We show that by monitoring the human behavior as well as the vehicle state, we can infer the human state and enhance the quality of decision making in a driver assistance system. The resulting control policy is obtained by solving an optimal planning problem of the proposed hidden mode hybrid system. The policy can automatically balance the decision making about when to give warning to the driver and when to actually intervene in the control of the vehicle.

Building on our previous work [1], in this paper we demonstrate how it is possible to improve flight control of a MAV that experiences aerodynamic disturbances caused by objects on its path. Predictions based on low resolution depth images taken at a distance are incorporated into the flight control loop on the throttle channel as this is adjusted to target undisrupted level flight. We demonstrate that a statistically significant improvement (p ≪ 0.001) is possible for some common obstacles such as boxes and steps, compared to using conventional feedback-only control. Our approach and results are encouraging toward more autonomous MAV exploration strategies.

Besides precise and dense geometric information, some LiDARs also provide intensity information and multiple echoes, information that can advantageously be exploited to enhance the performance of the purely geometric classification approaches. This information indeed depends on the physical nature of the perceived surfaces, and is not strongly impacted by the scene illumination - contrary to visual information. This article investigates how such information can augment the precision of a point cloud classifier. It presents an empirical evaluation of a low cost LiDAR, introduces features related to the intensity and multiple echoes and their use in a hierarchical classification scheme. Results on varied outdoor scenes are depicted, and show that more precise class identification can be achieved using the intensity and multiple echoes than when using only geometric features.

An automated diagnosis methodology is necessary for the maintenance of superannuated social infrastructures. In this context, the hammering test is an efficient inspection method, and it has been widely used because of the resulting accuracy and efficiency of operation. While robotic automation of the hammering inspection method is highly desirable, the development of an automatic diagnostic algorithm that can operate at actual inspection sites is essential. Furthermore, portability of the diagnostic algorithm is also highly desirable. In this study, in order to construct reliable detectors and to improve their portability for the performance of the hammering test, we propose a boosting-based defect detector that is robust against variations in environmental conditions. In particular, we present the construction of a noise-robust classifier with a refinement of the feature values extracted from hammering sounds and an updating rule of template vectors of its evaluation function. Our experimental results in a concrete tunnel demonstrate the effectiveness of the proposed method; the accuracy of the classifier at an actual site and adaptivity to environmental noise are confirmed.

Probabilistic Roadmap Methods (PRMs) solve the motion planing problem by constructing a roadmap (or graph) that models the motion space when feasible local motions exist. PRMs and variants contain several phases during roadmap generation i.e., sampling, connection, and query. Some work has been done to apply machine learning to the connection phase to decide which variant to employ, but it uses a global learning approach that is inefficient in heterogeneous situations. We present an algorithm that instead uses local learning: it only considers the performance history in the vicinity of the current connection attempt and uses this information to select good candidates for connection. It thus removes any need to explicitly partition the environment which is burdensome and typically difficult to do. Our results show that our method learns and adapts in heterogeneous environments, including a KUKA youBot with a fixed and mobile base. It finds solution paths faster for single and multi-query scenarios and builds roadmaps with better coverage and connectivity given a fixed amount of time in a wide variety of input problems. In all cases, our method outperforms the previous adaptive connection method and is comparable or better than the best individual method.

Multi-robot teams offer key advantages over single robots in exploration missions by increasing efficiency (explore larger areas), reducing risk (partial mission failure with robot failures), and enabling new data collection modes (multi-modal observations). However, coordinating multiple robots to achieve a system-level task is difficult, particularly if the task may change during the mission. In this work, we demonstrate how multiagent cooperative coevolutionary algorithms can develop successful control policies for dynamic and stochastic multi-robot exploration missions. We find that agents using difference evaluation functions (a technique that quantifies each individual agent's contribution to the team) provides superior system performance (up to 15%) compared to global evaluation functions and a hand-coded algorithm.

Tubular continuum robots are particularly useful for minimally invasive surgery due to their small size and ability to dexterously manipulate in constrained environments. The robot can be used teleoperated or operated (semi-)autonomously. In both cases, safe manipulation has to be ensured considering constraints from the patient's unique anatomy and pathology to prevent injury of nerves, vessels, and soft tissue. While probabilistic path planning methods have been proposed for tubular continuum robots which generate sequences of collision free configurations in obstructed environments, these methods do usually not explicitly consider different trajectory generations. In this paper, we evaluate four trajectory generation strategies in terms of the resulting Cartesian paths and spatial extent of the course of motion. We demonstrate the implication of the different strategies on both collision free path planning and an example medical scenario. In conclusion, consideration of trajectory generation strategies is indispensable, if the distance between intermediate configurations increases - even though those intermediate configurations might be collision free.

This paper introduces the control method for the impact force based on feed-forward control considering the stiffness ellipse of a biped robot. Assuming dynamic motion, such as jumping and running, the jumping direction depends on the dynamics in response to the ground reaction force. Therefore, it is necessary to control the ground reaction force in order to stabilize the system. However, control of the ground reaction force involves a fundamental problem: feedback control does not work properly against the impact force owing to the control bandwidth limitation. This study introduces a method for the stiffness ellipse control utilizing three antagonist pairs of six pneumatic artificial muscles. When the tip of the biped robot makes contact with the ground, an external force is induced in the direction that has lower stiffness. By utilizing this property of the stiffness ellipse, it is possible to control the ground reaction force by feed-forward control, which is not dependent on the control bandwidth. The results of several experiments convince us that the relationship between the ground reaction force and the stiffness ellipse is almost linear, and the ground reaction force can be controlled with high reproducibility by adjusting the stiffness ellipse. It is possible to stabilize the system during dynamic movements because this method can control the impact force.

In order to perform productive autonomous excavation of a fragmented rock pile, it is necessary to recognize the condition of the fragmented rock pile and to plan appropriate excavating motion according to the condition of the fragmented rock pile. In this paper, we propose imitation-based motion planning method, and develop a recognizer of rock pile condition and an excavating motion planner. We also develop an 1/10-scale excavation model and conduct excavation experiment. In the experiment, the proposed method works well from the view point of productivity. It is confirmed that the fragmented rock pile condition can be described by its shape and the particle size distribution of its surface. The proposed approach has feasibility in autonomous excavation of the fragmented rock pile. Experimental results also reveal that both the number of learning data and the diversity of learning data are important to realize a high-productive excavation.

This paper shows how to push an unknown object in the plane from an initial pose to a target pose with two cooperating mobile robots. On the object motion, we deliberately impose non-holonomic velocity constraint with pushing mobile robots. This yields smooth and efficient trajectories. Pushing manipulation is performed, for the first time, with a new uncalibrated image-based control scheme. This is achieved by transforming the image information to a scaled Euclidean space without using any metric information or calibration. Stability of the control law is also demonstrated.

The mass parameters of robots influence performances of model-based control and validation of the simulation results. The mass parameters provided by CAD data are usually rough approximation of the true parameters. Therefore several methods for estimation of those parameters have been proposed. Their precision depends on the used motion, called optimal exciting trajectories. This paper describes a new approach to determine humanoid robot exciting trajectories for mass parameters identification. The method was inspired by the studies done in the field of human mass parameters identification, and it is based on observation of condition numbers of sub-regressor matrices created from the columns of the regressor matrix. The method has been experimentally applied to identify mass parameters of HRP-2 and HRP-4 humanoid robots. The proposed method is able to reconstruct ground reaction forces and force moments more accurately than parameters obtained from CAD data.

In this study, we experimentally investigated the process leading to fracture in tofu grasping by deformable fingertips filled with a fluid. In our previous papers [1, 2], we developed deformable fingertips using a rubber bag filled with a viscoelastic fluid, and presented a strategy for delicate tofu grasping without any advance knowledge about fracture. However, the predication point was close to fracture, and the prediction was then still a gamble. In order to realize fracture prediction at an earlier stage, we examined the process leading to fracture when pushing tofu by the deformable fingertips. The stiffness of the fingertips can be controlled with the pressure of the fluid inside the fingertips. The pushing force and fluid pressure were examined for different levels of stiffness of the fingertips. The main findings and contributions are as follows. 1) The convergence of the ratio of the contact force to fluid pressure gives an indication of dent occurrence. This convergence could be seen when fingertip rubber bag was not filled (low stiffness). 2) It was easier for a dent to occur when the fingertip rubber bag was not filled than when it was filled (high stiffness). 3) Changes in the rate of increase of the fluid pressure as the tofu was pushed were repeatedly observed. We defined this as a phase change and present a method for detecting such changes. The phase change points were detected by comparing the fitting accuracies of different approximation models. 4) The last and second to the last phase changes before fracture were detected by detecting the first phase change (after the convergence of the rate of the contact force to fluid pressure if the fingertip bag was not completely filled). The detected points can be regarded as alert points indicating a fracture risk that is not close to the fracture point.

In this paper, a modified entropy-based algorithm is proposed for identification and reconstruction of a complex weld geometry. The edge of the weld geometry is identified based on minimizing a modified entropy-type cost function, and the weld geometry is reconstructed based on the detected edge. In addition, the volume of the weld geometry is computed using the point cloud samples of the identified weld geometry, and the effects of Gaussian noise are also considered. Our simulation results using the proposed reconstruction algorithm demonstrate efficient identification and reconstruction of a complex weld geometry in the presence of Gaussian noise.

This work introduces the IRON keypoint detector and the IRON descriptor which enable high-speed and high-accuracy alignment of 3D depth maps. Instead of using raw point values for storing 3D-scenes, all algorithms were designed to operate on Normal Distribution Transforms (NDT), since NDT-maps provide a highly memory-efficient representation of depth data. By taking into account surface curvature and object shape within NDT-maps, patches with strong surface variability can be recognized and described precisely. In this paper, the whole feature extraction process, as well as descriptor matching, outlier detection, and the final transform calculation between NDT-maps is elaborated. The presented technique is particularly insensitive to an initial offset between both maps, has a high robustness, and it achieves more than 75 NDT-map alignments per second (including complete memory allocation each time as well) in two large publicly available depth datasets while using only a single core of a modern Intel i7 CPU. Even though the main focus of this work was placed on the proposed IRON registration algorithm, two specific applications of this NDT-matching approach are outlined in the second part, namely robot pose tracking and NDT-one-shot localization within densely furnished domestic environments.

Fish caudal fin is a prominent example of biological propulsion, in which the caudal peduncle, fin rays and fin membrane together form a dynamic locomotory system. In this paper, we developed a bio-robotic model to mimic the caudal fin structure and kinematics of Bluegill Sunfish (Lepomis macrochirus). We coupled controlled oscillations in both heave and pitch directions to the robot to model the caudal peduncle motion of swimming fishes. Synchronized multi-axis force transducer and particle image velocimetry were then used to quantify the hydrodynamic forces and wake flow in the vertical plane. We found that changing the flow speed and the phase between the caudal fin ray and peduncle locomotion resulted in significantly different lift and thrust force. DPIV results showed that the vortex jet angle changed with flow speed and phase φ. According to current experiment, we hypothesized that the fish caudal fin may function as a “flexible vectoring propeller”, and may be responsible for the three-dimensional maneuverability of fishes.

We present a new hybrid approach to computing penetration depth (PD) for general polygonal models. Our approach exploits both local and global approaches to PD computation and can compute error-bounded PD approximations for both deep and shallow penetrations. We use a two-step formulation: the first step corresponds to a global approximation approach that samples the configuration space with bounded error using support vector machines; the second step corresponds to a local optimization that performs a projection operation refining the penetration depth. We have implemented this hybrid algorithm on a standard PC platform and tested its performance with various benchmarks. The experimental results show that our algorithm offers significant benefits over previously developed local-only and global-only methods used to compute the PD.

Mapping evolving environments requires an update mechanism to efficiently deal with dynamic objects. In this context, we propose a new approach to update maps pertaining to large-scale dynamic environments with semantics. While previous works mainly rely on large amount of observations, the proposed framework is able to build a stable representation with only two observations of the environment. To do this, scene understanding is used to detect dynamic objects and to recover the labels of the occluded parts of the scene through an inference process which takes into account both spatial context and a class occlusion model. Our method was evaluated on a database acquired at two different times with an interval of three years in a large dynamic outdoor environment. The results point out the ability to retrieve the hidden classes with a precision score of 0.98. The performances in term of localisation are also improved.

Traditional underwater vehicle-manipulator systems (UVMS) consist mainly of an underwater vehicle on which a manipulator is mounted. In addition to the vehicle thrusters, this paper proposes to use cables to actuate UVMSs. The main goal of this extra source of actuation is to enhance UVMS work capabilities. Similarly to cable-driven parallel robots, the cables are attached to the UVMS and a set of fixed winches permit the control of the cable lengths or tensions. We refer to this new system as a hybrid cable-thruster (HCT)-actuated UVMS. Based on the kinematic and dynamic model of HCTactuated UVMSs, this paper focuses on the determination of the available set of end-effector forces in a suitable mathematical representation. An illustrative simple planar example of HCTactuated UVMS is also presented.

Here we present a suite of theoretical, computational, and experimental studies culminating in the first aerial and aquatic capable insect-scale robot. We develop a computational fluid dynamics (CFD) simulation to model fluid-wing interaction in air and water. From CFD and a system dynamics analysis we predict that a multi-modal flapping strategy will enable locomotion in both air and water for a single device. We validate the CFD predictions by running at-scale, robotic wing-flapping experiments. Finally, we demonstrate for the first time a flying and swimming capable flapping-wing insect-like robot.