IROS 2016 Papers

We develop a novel approach for decision making under uncertainty in high-dimensional state spaces, considering both active unfocused and focused inference, where in the latter case reducing the uncertainty of only a subset of variables is of interest. State of the art approaches typically first calculate the posterior information (or covariance) matrix, followed by its determinant calculation, and do so separately for each candidate action. In contrast, using the generalized matrix determinant lemma, we avoid calculating these posteriors and determinants of large matrices. Furthermore, as our key contribution we introduce the concept of calculation re-use, performing a onetime computation that depends on state dimensionality and system sparsity, after which evaluating the impact of each candidate action no longer depends on state dimensionality. Such a concept is derived for both active focused and unfocused inference, leading to general, non-myopic and exact approaches that are faster by orders of magnitude compared to the state of the art. We verify our approach experimentally in two scenarios, sensor deployment (focused and unfocused) and measurement selection in visual SLAM, and show its superiority over standard techniques.

With the accelerated development of robot technologies, optimal control becomes one of the central themes of research. In traditional approaches, the controller, by its internal functionality, finds appropriate actions on the basis of the history of sensor values, guided by the goals, intentions, objectives, learning schemes, and so forth. While very successful with classical robots, these methods run into severe difficulties when applied to soft robots, a new field of robotics with large interest for human-robot interaction. We claim that a novel controller paradigm opens new perspective for this field. This paper applies a recently developed neuro controller with differential extrinsic synaptic plasticity to a muscle-tendon driven arm-shoulder system from the Myorobotics toolkit. In the experiments, we observe a vast variety of self-organized behavior patterns: when left alone, the arm realizes pseudo-random sequences of different poses. By applying physical forces, the system can be entrained into definite motion patterns like wiping a table. Most interestingly, after attaching an object, the controller gets in a functional resonance with the object's internal dynamics, starting to shake spontaneously bottles half-filled with water or sensitively driving an attached pendulum into a circular mode. When attached to the crank of a wheel the neural system independently develops to rotate it. In this way, the robot discovers affordances of objects its body is interacting with.

This contribution compares two approaches for applying disturbance observers (DOBs) to the torque control problem of series elastic actuators (SEAs). It is demonstrated that they are in fact equivalent for linear models in terms of their ability to reject disturbances and enforce nominal model dynamics. The closed loop and error transfer functions for the DOB-based approaches are compared to a fully linear plant and a nonlinear plant without DOB. Simulations demonstrate that the DOBs are able to increase the bandwidth of the nonlinear plant significantly, up to that of the linear plant. Furthermore, the DOBs significantly increase the tracking accuracy at low frequencies.

Multi-robot information gathering teams typically require communication for data fusion and cooperative decision making. However, when communication takes place over wireless networks, stringent bandwidth limits apply. These limits raise the need for efficient utilisation of available communication resources in a manner that balances information gathering utility with communication costs or limits. In our previous work, we introduced the dynamic information flow (DIF) problem as a general formulation of this trade-off. We introduced two variants of the problem addressing the issue of communication efficiency for data fusion only. In this paper, we extend one of the variants to address communication efficiency for both data fusion and cooperative decision making in a synergistic manner. We present a solution to this new variant that integrates a multi-cast routing algorithm with information structure optimisation. This solution allows teams that involve high-data-rate sensors and tight coordination to respect bandwidth limits. Through several simulations we verify that our solution significantly reduces bandwidth usage of such teams while maintaining information gathering performance.

Given the current capabilities of autonomous vehicles, one can easily imagine autonomy released on the road in the near future. However, it can be assumed that the transition will not be instantaneous, meaning they will have to be capable of driving well in a mixed environment, with both humans and other autonomous vehicles on the road. This leaves a number of concerns for autonomous vehicles in terms of dealing with human uncertainty and understanding of cooperation on the road. This work demonstrates the need for focusing on communication and collaboration between autonomy and human drivers. After analyzing how drivers perform cooperative maneuvers (e.g. lane changing), key cues were identified for conveying intent through nonverbal communication. It was found that human observers can predict lane changes with over two seconds in prior to the lane departure, without use of a turning signal. Building on this concept, an autonomous control scheme is proposed that aims to capture these subtle motions before executing a lane change. To compare the proposed human-inspired methods, three possible control schemes for autonomous vehicles are implemented for a validation study on human subjects to provide feedback on their experience. By properly conveying intent through nuanced trajectory planning, we show that drivers can predict the autonomous vehicle's actions with 40% increase in prediction time when compared to traditional control methods, both as a passenger and while observing the autonomous vehicle.

We address the problem of estimating an external wrench acting along the structure of a robot manipulator, together with the contact position where the external force is being applied. For this, we consider the combined use of a force/torque sensor mounted at the robot base and of a model-based virtual sensor. The virtual sensor is provided by the residual vector commonly used for collision detection and isolation in human-robot interaction. Integrating the two types of measurement tools provides an efficient way to estimate all unknown quantities, using also the recursive Newton-Euler algorithm for dynamic computations. Different operative conditions are considered, including the special cases of pointwise interaction (pure contact force), known contact location, and of a base sensor measuring only forces. We highlight also the conditions for a correct estimation to be fully virtual, i.e., without resorting to a force/torque sensor. Realistic simulations assess the estimation performance for a 7R robot in motion, subject to an unknown external force applied to an unknown location.

This paper deals with the problem of combining motion and task assignment for a dual-arm robotic system. Each arm of the system performs independent tasks in a cluttered environment. Robot actions are determined to remove potential obstacles and obtain collision-free paths to grasp the target objects. The approach uses the information provided by the motion planner to build a graph structure in order to represent the obstacles to be removed. The graph is used, first, to decide which is the next motion path to be computed, and, second, to assign the tasks to each arm of the robotic system. The approach has been implemented for a dual-arm robotic system and real experiments are presented in the paper.

This paper proposes a trajectory planning strategy to simultaneously address multiple surveillance objectives such as complete area coverage, periodic surveillance, and guaranteed detection of a rogue element attempting to exit the area of interest. For agents with identical sensing capability, the proposed strategy defines a time-varying multi-agent formation on a Lissajous curve, which completes the aforementioned tasks in finite time with collision free paths for all agents in a 2-D rectangular region. This obviates the need for sensing and communication for collision avoidance among agents. A sufficient upper limit on agent dimensions that ensures collision free motion is derived. An algorithm is developed for choosing the number of agents and optimal Lissajous curve for a prescribed rectangle and agents' sensing capability. The optimal Lissajous curve is chosen to minimize the time period for repeated coverage and maximize the upper bound on agent size. The proposed algorithm is validated through computer simulations and experiments using differential drive robots.

Flying and walking robots can use their complementary features in terms of viewpoint and payload capability to the best in a heterogeneous team. To this end, we present our online collaborative navigation framework for unknown and challenging terrain. The method leverages the flying robot's onboard monocular camera to create both a map of visual features for simultaneous localization and mapping and a dense representation of the environment as an elevation map. This shared knowledge from the flying platform enables the walking robot to localize itself against the global map, and plan a global path to the goal by interpreting the elevation map in terms of traversability. While following the planned path, the absolute pose corrections are fused with the legged state estimation and the elevation map is continuously updated with distance measurements from an onboard laser range sensor. This allows the legged robot to safely navigate towards the goal while taking into account any changes in the environment. In this setup, our approach is independent of external localization, relative observations between the robots, and does not require an initial guess about the pose of the robots. The presented methods are fully integrated and we demonstrate their capabilities in an experiment with a hexacopter and a quadrupedal robot.

Animals and robots alike face challenges of flipping-over as they move in complex terrain. Small insects like cockroaches can rapidly right themselves when upside down, yet small fast-running legged robots are much less capable of ground-based self-righting. Inspired by the discoid cockroach that opens its wings to push against the ground to self-right, we designed actuated wings for robot self-righting based on recently-developed rounded shells for obstacle traversal [1]. We measured the self-righting performance of a robot using these actuated wings, and systematically studied the effects and trade-offs of wing opening magnitude, speed, symmetry, and wing geometry. Our study provided a proof-of-concept that robots can take advantage of an existing body structure (rounded shell) in novel ways (as actuated wings) to serve new locomotor functions, analogous to biological exaptations [2]. Our results demonstrated that the robot self-rights dynamically, with active wing pushing followed by passive falling, and benefits from increasing kinetic energy by pushing faster and longer. Our experiments also showed that opening both wings asymmetrically increases righting probability at low wing opening magnitudes.

Solutions to robotic manipulation problems can be substantially improved through integrated task and motion planning. Existing approaches typically focus on satisfaction, finding a feasible solution, instead of optimization. We formulate large-scale robotic manipulation problems as multi-level optimization, incorporating task, action, and motion planning. We develop an integrated planning approach for solving this optimization problem and generating a combined motion plan for a robot to optimize a task-level objective. This approach utilizes a combinatorial search algorithm for task planning and incrementally exploits information from lower-level optimization to improve the high-level task plan. Empirical results show that this integrated approach not only significantly outperforms a traditional top-down approach in solution quality, but also avoids infeasible lower-level motion plans.

Robotic systems often have complex configuration spaces that, when poorly set, can cause failures. In this work we take advantage of the close synergy between user and robot in co-robotic systems to better diagnose and overcome configuration failures. We leverage users' understanding of the system to mark failures they observe while the system is in operation. A marked failure indicates that the robot either “did not do something when it should have” or “did something when it should not have”. The failure marking is coupled with an automated analysis approach that identifies code predicates involving configuration parameters that may be relevant to each failure type, ranks the parameters according to their potential to be associated with the failure, and suggests adjustments based on the run-time outcome of those predicates. We present the approach, its implementation, and a preliminary study on a configurable unmanned air system. The results show how the approach can successfully help diagnose and adjust faulty configuration parameters in co-robotic systems.

In this paper, a closed-loop strategy for controlling the transient regime of Marangoni flows and thus to selectively manipulate and separate micro-particles is reported. Marangoni flows were generated by an infrared laser as a heat source. Simulations of the transient regime of Marangoni flows were performed with COMSOL Multiphysics to analyze the temperature field and velocity profile. The convection cell dynamic growth was controlled with the laser beam exposure time. Then, selected particles, glass beads ranging from 150 up to 212 μm, were dragged by small convection flows without reaching undesired particles. The closed-loop control improved the manipulation speed and precision in comparison to manual manipulation.

Manufacturing process yield errors that change the kinematics of parallel robots. In this paper we describe how the constraint errors due to the manufacturing of the revolute axes can influence the inverse kinematic model of the Agile Eye, a Spherical Parallel Robot largely studied in the literature. Starting from the nominal inverse kinematic model a complete model to take into account geometric errors on the axes of the revolute joints is obtained in closed form. Numerical simulations compare the results from the nominal and real design to show relevant changes and to confirm the statistical nonlinear nature of the problem. Finally, an application to the elastodynamics is provided to verify the influence of the constraint errors on the frequency range.

Micromanipulation of individual biological cells, such as embryos during preimplantation genetic diagnosis (PGD), is a delicate and time-consuming task. Two major procedures in PGD include the rotation of the embryo to gain a more favorable position for zona breaching, and the extraction of a blastomere after an opening has been made in the zona pellucida. Rotation tracking of cleavage-stage embryos has not been reported given their lack of distinctive features. In this manuscript, a geometric model for partially determining the three dimensional (3-D) angular position of 2-cell embryos using two dimensional (2-D) microscopic brightfield images was derived and verified using a computer generated model. This model was then applied using computer vision algorithms on a rotating cleavage-stage mouse embryo, demonstrating partial 3-D rotation tracking. Furthermore, embryo micromanipulation tasks are typically performed manually using micropipettes. Technological advances have made automation of these tasks possible. This manuscript also presents computer vision algorithms for the segmentation and calibration of micropipettes. The calibration procedure allowed automated position control of the micropipettes without the need for real-time vision feedback using micropipette recognition algorithms, and effective position control was verified using semi-automated blastomere extraction experiments. This manuscript presents preliminary work towards the automation of cell manipulation procedures.

We consider the problem of sorting a densely cluttered pile of unknown objects using a robot. This yet unsolved problem is relevant in the robotic waste sorting business.

While numerous gaits in the horizontal regime (e.g. walking or running) have been defined for legged systems on level ground, no dynamically grounded definitions have been developed for dynamic vertical running. Gaits have clear implications to robotic control strategy, efficiency, and stability. However, while several climbing robotic systems have been described as achieving `running', the question of whether distinct dynamic gaits exist and what classifies these gaits has not been rigorously explored. In this paper, by applying definitions developed in the horizontal regime, we show evidence of three distinct gaits as well as discuss the implications of these gaits on the development of dynamic climbing systems.

Object detection is an essential task for autonomous robots operating in dynamic and changing environments. A robot should be able to detect objects in the presence of sensor noise that can be induced by changing lighting conditions for cameras and false depth readings for range sensors, especially RGB-D cameras. To tackle these challenges, we propose a novel adaptive fusion approach for object detection that learns weighting the predictions of different sensor modalities in an online manner. Our approach is based on a mixture of convolutional neural network (CNN) experts and incorporates multiple modalities including appearance, depth and motion. We test our method in extensive robot experiments, in which we detect people in a combined indoor and outdoor scenario from RGB-D data, and we demonstrate that our method can adapt to harsh lighting changes and severe camera motion blur. Furthermore, we present a new RGB-D dataset for people detection in mixed in- and outdoor environments, recorded with a mobile robot.

This paper is about underpinning long-term operations of fleets of vehicles using visual localisation. In particular it examines ways in which vehicles, considered as independent agents, can share, update and leverage each others' visual experiences in a mutually beneficial way. We draw on our previous work in Experience-based Navigation (EBN) [1], in which a visual map supporting multiple representations of the same place is built, yielding real-time localisation capability for a solitary vehicle. We now consider how any number of such agents might operate in concert via data sharing policies that are germane to the shared task of lifelong localisation. We rapidly construct considerable maps by the conjoining of work distributed to asynchronous processes, and share expertise amongst the team by the selective dispensing of mission-specific map contents. We demonstrate and evaluate our system against 100km of data collected in North Oxford over a period of a month featuring diverse deviation in appearance due to atmospheric, lighting, and structural dynamics. We show that our framework is capable of creating maps in a fraction of the time required by single-agent EBN, with no significant loss in localisation robustness, and is able to furnish robots on real-world forays with maps which require much less storage.