RSS 2016 论文列表

Our goal is to automate the understanding of natural hand-object manipulation by developing computer vision- based techniques. Our hypothesis is that it is necessary to model the grasp types of hands and the attributes of manipulated objects in order to accurately recognize manipulation actions. Specifically, we focus on recognizing hand grasp types, object attributes and actions from a single image within an unified model. First, we explore the contextual relationship between grasp types and object attributes, and show how that context can be used to boost the recognition of both grasp types and object attributes. Second, we propose to model actions with grasp types and object attributes based on the hypothesis that grasp types and object attributes contain complementary information for characterizing different actions. Our proposed action model outperforms traditional appearance-based models which are not designed to take into account semantic constraints such as grasp types or object attributes. Experiment results on public egocentric activities datasets strongly support our hypothesis.

Autonomous surface and underwater vehicles (ASVs and AUVs) are increasingly being used for persistent monitoring of ocean phenomena. Typically, these vehicles are deployed for long periods of time and must operate with limited energy budgets. As a result, there is increased interest in recent years on developing energy efficient motion plans for these vehicles that leverage the dynamics of the surrounding flow field. In this paper, we present a graph search based method to plan time and energy optimal paths in a flow field where the kinematic actuation constraints on the vehicles are captured in our cost functions. We also use tools from topological path planning to generate optimal paths in different homotopy classes, which facilitates simultaneous exploration of the environment. The proposed strategy is validated using analytical flow models for large scale ocean circulation and in experiments using an indoor laboratory testbed capable of creating flows with ocean-like features. We also present a Riemannian metric based approximation for these cost functions which provides an alternative method for computing time and energy optimal paths. The Riemannian approximation results in smoother trajectories in contrast to the graph based approach while requiring less computational time.

We consider the problem of assigning software pro- cesses (or tasks) to hardware processors in distributed robotics environments. We introduce the notion of a task variant, which supports the adaptation of software to specific hardware con- figurations. Task variants facilitate the trade-off of functional quality versus the requisite capacity and type of target execution processors. We formalise the problem of assigning task variants to processors as a mathematical model that incorporates typical con- straints found in robotics applications; the model is a constrained form of a multi-objective, multi-dimensional, multiple-choice knapsack problem. We propose and evaluate three different solution methods to the problem: constraint programming, a constructive greedy heuristic and a local search metaheuristic. Furthermore, we demonstrate the use of task variants in a real in- stance of a distributed interactive multi-agent navigation system, showing that our best solution method (constraint programming) improves the system’s quality of service, as compared to the local search metaheuristic, the greedy heuristic and a randomised solution, by an average of 16%, 41% and 56% respectively.

The sophisticated and intricate connection between bat morphology and flight capabilities makes it challenging to employ conventional flying robots to replicate the aerial locomotion of these creatures. In recent work, a bat inspired soft Micro Aerial Vehicle (MAV) called Bat Bot (B2) with five Degrees of Actuation (DoA) has been constructed to mimic the flight behavior of a biological bat. Major differences in structural topology resulted from this simpler kinematic complexity, and thus it is necessary to find the dimensions of B2’s structure and the behavior of its actuators such that the wingbeat cycle of B2 closely mimics that of a biological bat. The current work assumes the previously designed structure of B2 and presents a synergistic design approach to imitate the kinematic synergies of a biological bat. Recent findings have unveiled that the most dominant synergies in a biological bat could be combined to accurately represent the original kinematic movement, therefore simplifying its dimensional complexity. In this work, Principal Component Analysis (PCA) has been employed in order to extract dominant principal components of biological bat flight kinematics. Thereafter, first and second principal components are chosen to shape the parametric kinematics and actuator trajectories of B2 through finite state nonlinear constrained optimization. The method yields a robot mechanism that despite having a few DoAs, it possesses several biologically meaningful morphing specializations.

We propose a system for performing structural change detection in street-view videos captured by a vehicle- mounted monocular camera over time. Our approach is moti- vated by the need for more frequent and efficient updates in the large-scale maps used in autonomous vehicle navigation. Our method chains a multi-sensor fusion SLAM and fast dense 3D reconstruction pipeline, which provide coarsely registered image pairs to a deep deconvolutional network for pixel-wise change detection. To train and evaluate our network we introduce a new urban change detection dataset which is an order of magnitude larger than existing datasets and contains challenging changes due to seasonal and lighting variations. Our method outperforms existing literature on this dataset, which we make available to the community, and an existing panoramic change detection dataset, demonstrating its wide applicability.

Communication with robots is challenging, partly due to their differences from humans and the consequent discrepancy in people’s mental model of what robots can see, hear, or understand. Transparency mechanisms aim to mitigate this challenge by providing users with information about the robot’s internal processes. While most research in human-robot interaction aim towards natural transparency using human-like verbal and non-verbal behaviors, our work advocates for the use of visualization-based transparency. In this paper, we first present an end-to-end system that infers task commands that refer to objects or surfaces in everyday human environments, using Bayesian inference to combine scene understanding, pointing detection, and speech recognition. We characterize capabilities of this system through systematic tests with a corpus collected from people (N=5). Then we design human-like and visualization- based transparency mechanisms and evaluate them in a user study (N=20). The study demonstrates the effects of visualizations on the accuracy of people’s mental models, as well as their effectiveness and efficiency in communicating task commands.

Gaits are crucial to the performance of locomotors. However, it is often difficult to design effective gaits for complex locomotors. Geometric mechanics offers powerful gait design tools, but the utilities of these tools have been limited to systems with two joints. Using shape basis functions, it is possible to approximate the kinematics of complex locomotors using only two shape variables. As a result, the tools of geometric mechanics can be used to study complex locomotion in an intuitive way. The choice of shape basis functions plays an important role in determining gait kinematics, and therefore the performance of a locomotor. To find appropriate basis functions, we introduce the shape basis optimization algorithm, an algorithm that iteratively improve basis functions to find effective kinematic programs. Applying this algorithm to a snake robot resulted a novel gait, which improves its speed of swimming in granular materials.

Having many degrees of freedom is both a blessing and a curse. A mechanism with a large number of degrees of freedom can better comply to and therefore better move in complex environments. Yet, possessing many degrees of freedom is only an advantage if the system is capable of coordinating them to achieve desired goals in realtime. This work supports the belief that a middle layer of abstraction between conventional planning and control is needed to enable robust locomotion of articulated systems in complex terrains. The basis for this abstraction is the notion that a system’s shape can be used to capture joint- to-joint coupling and provide an intuitive set of controllable parameters that adapt the system to the environment in real time. This paper presents a generalizable framework that specifies desired shapes in terms of shape functions. We show how shape functions can be used to link low-level controllers to high-level planners in a compliant control framework that directly controls shape parameters. The resultant shape-based controllers produce behaviors that enable robots to robustly feel their way through unknown environments. This framework is applied to the control of two separate mechanisms, a snake-like and a hexapod robot.

For a given robot and a given task, this paper addresses questions about which modifications may be made to the robot’s suite of sensors without impacting the robot’s behavior in completing its task. Though this is an important design-time question, few principled methods exist for providing a definitive answer in general. Utilizing and extending the language of combinatorial filters, this paper aims to fill that lacuna by introducing theoretical tools for reasoning about sensors and representations of sensors. It introduces new representations for sensors and filters, exploring the relationship between those elements and the specific information needed to perform a task. It then shows how these tools can be used to algorithmically answer questions about changes to a robot’s sensor suite. The paper substantially expands the expressiveness of combinatorial filters so that, where they were previously limited to quite simple sensors, our richer filters are able to reasonably model a much broader variety of real devices. We have implemented the proposed algorithms, and describe their application to an example instance involving a series of simplifications to the sensors of a specific, widely deployed mobile robot.

Perception of transparent objects has been an open challenge in robotics despite advances in sensors and data- driven learning approaches. In this paper, we introduce a new approach that combines recent advances in learnt object detectors with perceptual grouping in 2D, and projective geometry of apparent contours in 3D. We train a state of the art structured edge detector on an annotated set of foreground glassware. We assume that we deal with surfaces of revolution (SOR) and apply perceptual symmetry grouping in a 2D spherical transformation of the image to obtain a 2D detection of the glassware object and a hypothesis about its 2D axis. Rather than stopping at a single view detection, we ultimately want to reconstruct the 3D shape of the object and its 3D pose to allow for a robot to grasp it. Using two views allows us to decouple the 3D axis localization from the shape estimation. We develop a parametrization that uniquely relates the shape reconstruction of SOR to given a set of contour points and tangents. Finally, we provide the first annotated dataset for 2D detection, 3D pose and 3D shape of glassware and we show results comparable to category-based detection and localization of opaque objects without any training on the object shape.

Safe control of dynamical systems that satisfy tem- poral invariants expressing various safety properties is a challeng- ing problem that has drawn the attention of many researchers. However, making the assumption that such temporal properties are deterministic is far from the reality. For example, a robotic system might employ a camera sensor and a machine learned system to identify obstacles. Consequently, the safety properties the controller has to satisfy, will be a function of the sensor data and the associated classifier. We propose a framework for achieving safe control. At the heart of our approach is the new Probabilistic Signal Temporal Logic (PrSTL), an expressive language to define stochastic properties, and enforce probabilistic guarantees on them. We also present an efficient algorithm to reason about safe controllers given the constraints derived from the PrSTL specification. One of the key distinguishing features of PrSTL is that the encoded logic is adaptive and changes as the system encounters additional data and updates its beliefs about the latent random variables that define the safety properties. We demonstrate our approach by deriving safe control of quadrotors and autonomous vehicles in dynamic environments.

This paper considers the problem of routing and rebalancing a shared fleet of autonomous (i.e., self-driving) vehicles providing on-demand mobility within a capacitated trans- portation network, where congestion might disrupt throughput. We model the problem within a network flow framework and show that under relatively mild assumptions the rebalancing vehicles, if properly coordinated, do not lead to an increase in congestion (in stark contrast to common belief). From an algorithmic standpoint, such theoretical insight suggests that the problem of routing customers and rebalancing vehicles can be decoupled, which leads to a computationally-efficient routing and rebalancing algorithm for the autonomous vehicles. Numerical experiments and case studies corroborate our theoretical insights and show that the proposed algorithm outperforms state-of-the- art point-to-point methods by avoiding excess congestion on the road. Collectively, this paper provides a rigorous approach to the problem of congestion-aware, system-wide coordination of autonomously driving vehicles, and to the characterization of the sustainability of such robotic systems.

In this study, we present a framework for phase- space planning and control of agile bipedal locomotion while robustly tracking a set of non-periodic keyframes. By using a reduced-order model, we formulate a hybrid planning framework where the center-of-mass motion is constrained to a general sur- face manifold. This framework also proposes phase-space bundles to characterize robustness and a robust hybrid automaton to effectively design planning algorithms. A newly defined phase- space locomotion manifold is used as a Riemannian metric to measure the distance between the disturbed state and the planned manifold. Based on this metric, a dynamic programming based hybrid controller is introduced to produce robust locomotions. The robustness of the proposed framework is validated by using simulations of rough terrain locomotion recovery from external disturbances. Additionally, the agility of this framework is demonstrated by using simulations of the dynamic locomotion over random rough terrains.

Loop closure detection is an essential component for simultaneously localization and mapping in a variety of robotics applications. One of the most challenging problems is to perform long-term place recognition with strong perceptual aliasing and appearance variations due to changes of illumination, vegetation, weather, etc. To address this challenge, we propose a novel Robust Multimodal Sequence-based (ROMS) method for long-term loop closure detection, by formulating image sequence matching as an optimization problem regularized by structured sparsity-inducing norms. Our method is able to model the sparsity nature of place recognition, i.e., the current location should match only a small subset of previously visited places, as well as to model underlying structures of image sequences and incorporate multiple feature modalities to construct a discriminative scene representation. In addition, a new optimization algorithm is developed to efficiently solve the formulated problem, which has a theoretical guarantee to converge to the global optimal solution. To evaluate the ROMS algorithm, extensive experiments are performed using large-scale benchmark datasets, including St Lucia, CMU-VL, and Nordland datasets. Experimental results have validated that our algorithm outperforms previous loop closure detection methods, and obtains the state-of-the-art performance on long-term place recognition.

We conducted a study to investigate trust in and dependence upon robotic decision support among nurses and doctors on a labor and delivery floor. There is evidence that suggestions provided by embodied agents engender inappropriate degrees of trust and reliance among humans. This concern is a critical barrier that must be addressed before fielding intelligent hospital service robots that take initiative to coordinate patient care. Our experiment was conducted with nurses and physicians, and evaluated the subjects’ levels of trust in and dependence on high- and low-quality recommendations issued by robotic versus computer-based decision support. The support, generated through action-driven learning from expert demonstration, was shown to produce high-quality recommendations that were ac- cepted by nurses and physicians at a compliance rate of 90%. Rates of Type I and Type II errors were comparable between robotic and computer-based decision support. Furthermore, em- bodiment appeared to benefit performance, as indicated by a higher degree of appropriate dependence after the quality of recommendations changed over the course of the experiment. These results support the notion that a robotic assistant may be able to safely and effectively assist in patient care. Finally, we conducted a pilot demonstration in which a robot assisted resource nurses on a labor and delivery floor at a tertiary care center.

We describe a process that constructs robot-specific circuitry for motion planning, capable of generating motion plans approximately three orders of magnitude faster than existing methods. Our method is based on building collision detection circuits for a probabilistic roadmap. Collision detection for the roadmap edges is completely parallelized, so that the time to determine which edges are in collision is independent of the number of edges. We demonstrate planning using a 6-degree- of-freedom robot arm in less than 1 millisecond.

We present a framework for representing scenarios with complex object interactions, in which a robot cannot directly interact with the object it wishes to control, but must instead do so via intermediate objects. For example, a robot learning to drive a car can only indirectly change its pose, by rotating the steering wheel. We formalize such complex interactions as chains of Markov decision processes and show how they can be learned and used for control. We describe two systems in which a robot uses learning from demonstration to achieve indirect control: playing a computer game, and using a hot water dispenser to heat a cup of water.

This paper provides a new fully-decentralized al- gorithm for Collaborative Localization based on the extended Kalman filter. The major challenge in decentralized collaborative localization is to track inter-robot dependencies – which is particularly difficult in situations where sustained synchronous communication between robots cannot be guaranteed. Current approaches suffer from the need for particular communica- tion schemes, extensive bookkeeping of measurements, overly- conservative assumptions, or the restriction to specific mea- surement models. To the best of our knowledge, the algorithm we propose in this paper is the first one that tracks inter- robot correlations while fulfilling all of the following relevant conditions: communication is limited to two robots that obtain a relative measurement, the algorithm is recursive in the sense that it does not require storage of measurements and each robot maintains only the latest estimate of its own pose, and it supports generic measurement models. These particularly hard conditions make the approach applicable to a wide range of multi-robot applications. Extensive experiments carried out using real world datasets demonstrate the improved performance of our method compared to several existing approaches.

In recent years, the topic of multi-robot systems has become very popular. These systems have been demonstrated in various applications, including exploration, construction, and warehouse operations. In order for the whole system to function properly, sensor calibrations such as determining the camera frame relative to the IMU frame are important. Compared to the traditional hand-eye & robot-world calibration, a relatively new problem called the AX B = Y C Z calibration problem arises in the multi-robot scenario, where A, B, C are rigid body transformations measured from sensors and X,Y,Z are unknown transformations to be calibrated. Several solvers have been proposed previously in different application areas that can solve for X, Y and Z simultaneously. However, all of the solvers assume a priori knowledge of the exact correspondence among the data streams {Ai}, {Bi} and {Ci}. While that assumption may be justified in some scenarios, in the application domain of multi-robot systems, which may use ad hoc and asynchronous communication protocols, knowledge of this correspondence gen- erally cannot be assumed. Moreover, the existing methods in the literature require good initial estimates that are not always easy or possible to obtain. In this paper, we propose two probabilistic approaches that can solve the AXB = Y CZ problem without a priori knowledge of the correspondence of the data. In addition, no initial estimates are required for recovering X, Y and Z. These methods are particularly well suited for multi-robot systems, and also apply to other areas of robotics in which AXB = YCZ arises.