IROS 2017 论文列表

Researchers in the simultaneous localization and mapping (SLAM) community have taken for granted that uncertainty associated with the robot pose increases until the loop is closed. However, recently identified by [1], the monotonicity of uncertainty during exploration breaks when the robot returns to the initial position. In this paper, we propose a hypothesis that the monotonicity of pose uncertainty is preserved when the uncertainty is propagated on Lie groups rather than on Euclidean vector space. After deriving covariance propagated over Lie groups and Euclidean vector space, respectively, the monotonicity of uncertainty in each case is thoroughly investigated. Experiments with simulated and real-world scenarios on dead-reckoning validate our hypothesis on the monotonicity of uncertainty.

We aim to reliably predict whether a grasp on a known object is successful before it is executed in the real world. There is an entire suite of grasp metrics that has already been developed which rely on precisely known contact points between object and hand. However, it remains unclear whether and how they may be combined into a general purpose grasp stability predictor. In this paper, we analyze these questions by leveraging a large scale database of simulated grasps on a wide variety of objects. For each grasp, we compute the value of seven metrics. Each grasp is annotated by human subjects with ground truth stability labels. Given this data set, we train several classification methods to find out whether there is some underlying, non-trivial structure in the data that is difficult to model manually but can be learned. Quantitative and qualitative results show the complexity of the prediction problem. We found that a good prediction performance critically depends on using a combination of metrics as input features. Furthermore, non-parametric and non-linear classifiers best capture the structure in the data.

Autonomous mobile robots require motion planning algorithms that match limitations of on-board computing resources to safely navigate complex environments. In situations where near-optimality is preferential to runtime performance, search spaces that optimize their local connectivity to improve the global optimality of generated solutions are desirable. However, not all nodes in the search space benefit equally from optimization which can result in an inefficient use of computational resources and unnecessary increase in runtime. To address this limitation, we propose an approach called the Selectively Adaptive State Lattice which uses a heuristic based on the local environment to selectively perform optimization and obtain a balance between runtime performance and relative optimality. We present a statistical evaluation of local connectivity optimization and global search with the State Lattice, Adaptive State Lattice, and Selectively Adaptive State Lattice algorithms in randomly generated obstacle fields. We further highlight the performance of each method on a Clearpath Robotics TurtleBot2 in a qualitative physical experiment.

Tendon-driven continuum robots possess versatile application capabilities and have a robust design. The actuation of such robots with non-straight tendons that wrap around the backbone, described by a variable function, offers a lot of untapped potentials. While it has been shown that these continuum robots are able to take up complex shapes using only one actuated segment, the merits of non-straight tendon routing have not been quantified in terms of workspace and motion. In this paper, we show that one additional helically routed tendon can greatly benefit the robot's reachable workspace. For instance, the reachable workspace of a one-segment robot with 3 conventional straight tendons increases by 400 % by adding one helically routed tendon. Furthermore, the dexterity of such a continuum robot is improved, i.e. motion sequences to avoid obstacles or to twine an object for grasping. For the first time, the potential of tendon-driven continuum robots with two segments and helically routed tendons is investigated. The general findings on the merits of helical tendon routing are supported with both simulation and experimental results.

This paper presents a method to maximize the manipulability index of a redundant industrial manipulator while solving the Inverse Kinematics (IK) problem as an optimization problem. Even though the IK problem is a widely studied topic, the integration of the manipulability index into IK has rarely been taken into account. As the relation between this index and the joint variables of the robot is not straightforward, we have tested different formulations using approximated derivatives. Obstacle avoidance has also been considered and the effect of modifying the Cartesian trajectory during the execution of a task has been thoroughly analyzed. Different scenarios have been conducted in simulation and have proven that our modified inverse kinematics solver is efficient to maximize the manipulability index, even with the additional constraint of obstacle avoidance. Moreover, the relaxation of the trajectory constraints leads to a greater manipulability while ensuring the motion smoothness and satisfying the robot physical limitations.

In this paper, we address a generalization of the Orienteering Problem (OP) for curvature-constrained vehicles and to problems where it is allowed to collect a reward associated to each target location within a specified distance from the target. The addressed problem combines challenges of the combinatorial optimization of the OP (to select the most rewarding targets and find the optimal sequence to visit them) with the continuous optimization related to the determination of the waypoint locations and suitable headings at the waypoints for the considered Dubins vehicle such that the curvature-constrained path does not exceed the given travel budget and the sum of the collected rewards is maximized. The proposed generalization is called the Close Enough Dubins Orienteering Problem (CEDOP) and novel unsupervised learning approach is proposed to address computational requirements of this challenging planning problem. Based on the presented results, the proposed approach is feasible and provides a bit worse solution of CEDOP than the existing combinatorial approach but with significantly lower computational requirements.

We present an Omnidirectional Visual-Inertial Odometry (OVIO) approach based on Multi-State Constraint Kalman Filtering (MSCKF) to estimate the ego-motion of a moving platform. Instead of considering visual measurements on image plane, we use individual planes for each point that are tangent to the unit sphere and normal to the corresponding measurement ray. This way, we combine spherical images captured by omnidirectional camera with inertial measurements within the filtering method MSCKF. The key hypothesis of OVIO is that a wider field of view allows incorporating more visual features from the surrounding environment, thereby improving the accuracy and robustness of the motion estimation. Moreover, by using an omnidirectional camera, it is less likely to end up in a situation where there is not enough texture. We provide an evaluation of OVIO using synthetic and real video sequences captured by a fish-eye camera, and compare the performance with MSCKF using a perspective camera. The results show the superior performance of the proposed OVIO.

In this paper, we introduce a novel approach of reducing the communication load in distributed model predictive control (DMPC) for mobile robots. The key idea is to project the predicted state trajectory onto a grid resulting in an occupancy grid prediction. This approach has the advantage of utilizing continuous optimization methods while only quantized information is exchanged. We consider non-holonomic mobile robots to numerically and experimentally investigate the proposed method.

We present a new method to detect object affordances in real-world scenes using deep Convolutional Neural Networks (CNN), an object detector and dense Conditional Random Fields (CRF). Our system first trains an object detector to generate bounding box candidates from the images. A deep CNN is then used to learn the depth features from these bounding boxes. Finally, these feature maps are post-processed with dense CRF to improve the prediction along class boundaries. The experimental results on our new challenging dataset show that the proposed approach outperforms recent state-of-the-art methods by a substantial margin. Furthermore, from the detected affordances we introduce a grasping method that is robust to noisy data. We demonstrate the effectiveness of our framework on the full-size humanoid robot WALK-MAN using different objects in real-world scenarios.

Robot interaction planning is a computationally expensive process which rarely makes use of previous experiences in a deliberative manner. This paper addresses this issue by examining dimensionality reduction techniques to allow comparison of objects in a robot's environment based on the way they react to robot manipulation. We compare a number of techniques which can map objects from an observation space, which may contain thousands of dimensions, to a lower dimensionality space - the embedding space - which allows objects to be compared in an efficient manner, making knowledge transfer between similar objects more computationally tractable.

Achieving object exploration with passive vision and active touch has been under investigation for thirty years. We build upon recent progress in biomimetic active touch that combines perception via Bayesian evidence accumulation with controlling the tactile sensor using perceived stimulus location. Here, passive vision is combined with active touch by providing a visual prior for each perceptual decision, with the precision of this prior setting the relative contribution of each modality. The performance is examined on an edge following task using a tactile fingertip (the TacTip) mounted on a robot arm. We find that the quality of exploration is a U-shaped function of the relative contribution of vision and touch; moreover, multi-modal performance is more robust, completing the contour when touch alone fails. The overall system has several parallels with biological theories of perception, and thus plausibly represents a robot model of visuo-tactile exploration in humans.

We study the HRP-2Kai humanoid robot's ability to conduct the precise industrial task of fastening bolts in aircraft production. Our contribution stands mainly in high integration of different modules within the whole-body Quadratic Programing-based controller that has not been yet confronted to tasks demanding high precision in the execution and tuning. This includes the use of a robust visual servoing algorithm which allows the robot to move autonomously to a desired target and the design of specific tasks and estimators: a learning and admittance control that enables the robot to interact smoothly with its environment, and a fast and safe method to autonomously detect correct tool on nut insertion. We then show that the controller indeed enables our humanoid robot to achieve such a high precision task.

In SLAM, the size of the state vector tends to grow when exploring unknown environments, causing the ever-increasing computational complexity. To reduce the computational cost, one needs to continuously marginalize part of previous states (features and/or poses). This can be achieved by using either the null-space operation or Schur complement based marginalization. The former was originally developed particularly for visual-inertial odometry (VIO), while the latter is the standard approach to reduce the state vector in bundle adjustment (BA). In this paper, for the first time ever, we prove that under mild assumptions (i.i.d. white Gaussian noise model and same linearization points) these two techniques retain the same amount of information about the state, which is validated with real-world experiments. Moreover, based on this key insight, we derive analytically the left null-space expressions for multi-state constraint Kalman filter (MSCKF)-based VIO, which is verified through Monte-Carlo simulations.

Stroke rehabilitation technologies have focused on reducing treatment cost while improving effectiveness. Rehabilitation robots are generally developed for home and clinical usage to: 1) deliver repetitive practice to post-stroke patients, 2) minimize therapist interventions, and 3) increase the number of patients per therapist, thereby decreasing the associated cost. The control of rehabilitation robots is often limited to black-or gray-box approaches; thus, safety issues regarding the human-robot interaction are not easily considered. To overcome this issue, controllers working with physics-based models gain more importance. In this study, we have developed an efficient two dimensional (2D) human-robot interaction model to implement a model-based controller on a planar end-effector-type rehabilitation robot. The developed model was used within a nonlinear model predictive control (NMPC) structure to control the rehabilitation robot. The GPOPS-II optimal control package was used to implement the proposed NMPC structure. The controller performance was evaluated by simulating the human-robot rehabilitation system, modeled in MapleSim®. In this system, a musculoskeletal model of the arm interacting with the robot is used to predict movement and muscle activation patterns, which are used by the controller to provide optimal assistance to the patient. In simulations, the controller achieved desired performance and predicted muscular activities of the dysfunctional subject with a good accuracy. In our future work, a structure exploiting the NMPC framework will be developed for the real-time control of the rehabilitation robot.

This paper extends the previously proposed Explicit Force Controller based on Time Domain Passivity Approach. When using the classical passivity observer, we encounter an energy accumulation problem: if the system stays a long time in stable contact, energy is dissipated and the passivity observer builds up a large value. This causes the passivity controller to be triggered late after the interaction becomes unstable, so we lose the advantage of the passivity controller. In order to deal with this energy accumulation issue, we propose two new passivity observers that allow us to quickly detect potential instability despite the accumulated energy. We prove the theoretical validity of these new observers. In addition, we propose a more generalized way of implementing the passivity based explicit force controller on a multi-DoF manipulator, using a model for the robot and environment that includes sensor flexibility, and a hybrid position/force controller in the operational space framework. The proposed method is experimentally tested with KUKA IIWA, and the improved performance is verified.

In order to be able to perform complex and arduous tasks, stiffness articular identification of industrial robots is a current approach to predict the deflection under static or dynamic loading. Manufacturers propose new features to take the loading into account and a new generation of industrial robot equiped with double encoding systems are proposed. However, current methods brings some drawbacks when the ratio between the stiffness arm and the wrist one is too high. In this paper, we propose a new approach to take this aspect into account by decoupling the arm identification and the wrist one. We compare then our method regarding two current methods and applied it on this new industrial robot. The results highlight the stability and the quality of the stiffness articular estimation with and without activating the double encoding system. On our data, we are able to take into account 84% of the global deflection.

State-of-the-art personal robots must perform complex manipulation tasks to be viable in assistive scenarios. However, many of these robots, like the PR2, use manipulators with high degrees-of-freedom. The complexity of these robots lead to large dimensional state spaces, which are difficult to fully explore. Our previous work introduced the IDRRL algorithm, which compresses the learning space by transforming a high-dimensional learning space onto a lower-dimensional manifold while preserving expressivity. In this work we formally prove that IDRRL maintains PAC-MDP guarantees. We then improve upon our previous formulation of IDRRL by introducing cascading autoencoders (CAE) for dimensionality reduction, producing the new algorithm IDRRL-CAE. We demonstrate the improvement of this extension over our previous formulation, IDRRL-PCA, in the Mountain Car and Swimmers domains.

Model-based control of dynamical systems typically requires accurate domain-specific knowledge and specifications of possibly proprietary system components. In the context of autonomous driving, steering actuator dynamics can be difficult to model due to an integrated proprietary power steering control module. While first-principles models derived from physics laws can often approximate the system behavior, it remains generally difficult to capture non-physically derived behavior based on proprietary software algorithms in the power steering system. To overcome this limitation, this work instead employs a recurring neural network to model the steering dynamics of an autonomous vehicle. The resulting model is then integrated into a Nonlinear Model Predictive Control scheme to generate feedforward steering commands for embedded control. The proposed approach is compared to traditional first-principles steering modeling through on-vehicle experiments and statistical data validation. As a result, it is shown that the neural network model can be automatically generated with less domain-specific knowledge, can predict steering dynamics more accurately, and perform comparably to a high-fidelity first principles model when used for controlling the steering system of a self-driving vehicle.

In this paper, we presented a high-power, self-balancing, passively and software-controlled active compliant, and wearable hip exoskeleton to provide walking and balance assistance. The device features powered hip abduction/adduction (HAA) and hip flexion/extension (HFE) modules to provide assistance in both sagittal and frontal planes. Each module's actuation unit employs a Series Elastic Actuator (SEA) to achieve passive compliance. The hip exoskeleton can work in two basic operation modes: human-in-charge and robot-in-charge. Both modes are integrated into the low-level controller based on the admittance control, making transitions smooth and stable. A new balance controller based on the “extrapolated center of mass” (XCoM) concept is presented for real-time control hip abduction/adduction to keep the center of mass (CoM) within the support polygon. The exoskeleton controller is designed to encourage participation in walking instead of overriding users' intrinsic behavior to achieve effective assistance and training. Our preliminary experiments on a healthy subject using the hip exoskeleton demonstrated the potential effectiveness of the device and controller in assisting locomotion.