IROS 2017 Papers

Performing manipulation with robotic arms requires a method for planning trajectories that takes multiple factors into account: collisions, joint limits, orientation constraints, torques, and duration of a trajectory. We present an approach to efficiently optimize arm trajectories with respect to multiple criteria. Our work extends Stochastic Trajectory Optimization for Motion Planning (STOMP). We optimize trajectory duration by including velocity into the optimization. We propose an efficient cost function with normalized components, which allows prioritizing components depending on user-specified requirements. Optimization is done in two stages: first with a partial cost function and in the second stage with full costs. We compare our method to state-of-the art methods. In addition, we perform experiments on real robots: centaur-like robot Momaro and an industrial manipulator.

This paper presents an accurate and computationally efficient stairway detection algorithm, robust to measurement noise. While most similar methods depend heavily on detecting three-dimensional (3D) features, this work focuses on locating two-dimensional (2D) features embedded in 3D data. The main idea is: each stair is a sequence of lines with similar domain of definition, and the staircase is a sequence of such elements. To avoid unnecessary computations, an octree is used to down-sample the environment point-cloud data. The resulting centroids are stored in n arrays, each containing points with equal height. This effectively segments the 3D data into n 2D slices. The Hough transform is used to scan each array for lines. the remainder of the algorithm detects lines with the same domain of definition and determines if sequences that may correspond to stairways exist. If these sequences meet the required tolerances, then a staircase is detected. For each detected stairway a model that contains the width, height, depth, and amount of stairs is returned. The proposed algorithm is able to efficiently locate and model staircases on different poses and is robust to noise from the measured data, as indicated by the experimental results.

We introduce a simple yet effective sampling-based planner that is tailored for bottleneck pathfinding: Given an implicitly-defined cost map M : Rd Å R, which assigns to every point in space a real value, we wish to find a path connecting two given points, which minimizes the maximal value with respect to M. We demonstrate the capabilities of our algorithm, which we call bottleneck tree (BTT), on several challenging instances of the problem involving multiple agents, where it outperforms the state-of-the-art cost-map planning technique T-RRT. In addition to its efficiency, BTT requires the tuning of only a single parameter: the number of samples. On the theoretical side, we study the asymptotic properties of our method and consider the special setting where the computed trajectories must be monotone in all coordinates. This constraint arises in cases where the problem involves the coordination of multiple agents that are restricted to forward motions along predefined paths.

Robotic lawn-mowers are required to stay within a predefined working area, otherwise they may drive into a pond or on the street. This turns navigation and path planning into safety critical components. If we consider using SLAM techniques in that context, we must be able to provide safety guarantees in the presence of sensor/actuator noise and featureless areas in the environment. In this paper, we tackle the problem of planning a path that maximizes robot safety while navigating inside the working area and under the constraints of limited computing resources and cheap sensors. Our approach uses a map of the environment to estimate localizability at all locations, and it uses these estimates to search for a path from start to goal in belief space using an extended heuristic search algorithm. We implemented our approach using C++ and ROS and thoroughly tested it on simulation data recorded on eight different gardens, as well as on a real robot. The experiments presented in this paper show that our approach leads to short computation times and short paths while maximizing robot safety under certain assumptions.

A model of hierarchical Slow Feature Analysis (SFA) enables a mobile robot to learn a spatial representation of its environment directly from images captured during a random walk. After the unsupervised learning phase a subset of the resulting representations are orientation invariant and code for the position of the robot. Hence, they change monotonically over space even though the variation of the sensory signals received from the environment might change drastically e.g. during rotation on the spot. Furthermore, the property of spatial smoothness allows us to infer a navigation direction by taking the difference between the measurement at the current location and a measurement at a target location. In our work we investigated the use of slow feature representations, learned for a specific environment, for the purpose of navigation. We present a straightforward method for navigation using gradient descent on the difference between two points specified in slow feature space. Due to its slowness objective, the resulting slow feature representations implicitly encode information about static obstacles, allowing a mobile robot to efficiently circumnavigate them by simply following the steepest gradient in slow feature space.

This paper addresses the complete area coverage problem of a known environment by multiple-robots. Complete area coverage is the problem of moving an end-effector over all available space while avoiding existing obstacles. In such tasks, using multiple robots can increase the efficiency of the area coverage in terms of minimizing the operational time and increase the robustness in the face of robot attrition. Unfortunately, the problem of finding an optimal solution for such an area coverage problem with multiple robots is known to be NP-complete. In this paper we present two approximation heuristics for solving the multi-robot coverage problem. The first solution presented is a direct extension of an efficient single robot area coverage algorithm, based on an exact cellular decomposition. The second algorithm is a greedy approach that divides the area into equal regions and applies an efficient single-robot coverage algorithm to each region. We present experimental results for two algorithms. Results indicate that our approaches provide good coverage distribution between robots and minimize the workload per robot, meanwhile ensuring complete coverage of the area.

In this paper we propose a method for monocular visual-inertial odometry that utilizes image edges as measurements. In contrast to previous feature-based approaches, the proposed method does not employ any assumption on the geometry of the scene (e.g., it does not assume straight lines). It can thus use measurements from all image areas with significant gradient, similarly to direct semi-dense methods. However, in contrast to direct semi-dense approaches, the proposed method's measurement model is invariant to linear changes in the image intensity. The novel edge parameterization and measurement model we propose explicitly account for the fact that edge points can only provide useful information in the direction of the image gradient. We present both Monte-Carlo simulations, as well as results from real-world experimental testing, which demonstrate that the proposed edge-based approach to visual-inertial odometry is consistent, and outperforms the point-based one.

The ability to manipulate objects is the primary purpose of any robotic hand. However, when executing a grasp, the object and fingers rarely move exactly as planned. These unobserved deviations in the pose of the object or the contact configuration can make it impossible to solve a given task. In this paper, we presents a new approach to estimate the state of the grasp using only position and torque measurements from the joints of the hand. Based on the popular extended Kalman filter framework, the algorithm estimates the pose of the manipulated object, as well as the contact forces and positions on the surface of the object. It is able to reliably detect new and the loss of contacts and to incorporate this information in the estimation filter. The validity of this new method is shown in different grasp and manipulation tasks using David, a humanoid platform of DLR.

Learning and recognition is a fundamental process performed in many robot operations such as mapping and localization. The majority of approaches share some common characteristics, such as attempting to extract salient features, landmarks or signatures, and growth in data storage and computational requirements as the size of the environment increases. In biological systems, spatial encoding in the brain is definitively known to be performed using a fixed-size neural encoding framework - the place, head-direction and grid cells found in the mammalian hippocampus and entorhinal cortex. Particularly paradoxically, one of the main encoding centers - the grid cells - represents the world using a highly aliased, repetitive encoding structure where one neuron represents an unbounded number of places in the world. Inspired by this system, in this paper we invert the normal approach used in forming mapping and localization algorithms, by developing a novel place recognition algorithm that seeks out and leverages repetitive, mutually complementary landmark frequencies in the world. The combinatorial encoding capacity of multiple different frequencies enables not only the ability to achieve efficient data storage, but also the potential for sub-linear storage growth in a learning and recall system. Using both ground-based and aerial camera datasets, we demonstrate the system finding and utilizing these frequencies to achieve successful place recognition, and discuss how this approach might scale to arbitrarily large global datasets and dimensions.

In this work we present a co-located task space sensing and control system designed to control the end-effector motion of a mobile manipulator in the presence of dynamic and unknown base motion. We present a method for generating end-effector motion estimates at 1 kilohertz for use in real time control through visual-inertial sensor fusion. We show that a Moving Horizon Estimator outperforms Kalman filter-based methods in generating accurate predictive estimates for use in real time. We use this estimator to close a task space control loop directly at the end-effector, assuming no prior knowledge of the base pose and motion. We demonstrate the performance of this system on a hydraulically actuated arm which performs task-space tracking tasks in the presence of significant unknown base motion.

We present a real-time pattern generator for dynamic walking over rough terrains. Our method automatically finds step durations, a critical issue over rough terrains where they depend on terrain topology. To achieve this level of generality, we consider a Floating-base Inverted Pendulum (FIP) model where the center of mass can translate freely and the zero-tilting moment point is allowed to leave the contact surface. This model is equivalent to a linear inverted pendulum with variable center-of-mass height, but its equations of motion remain linear. Our solution then follows three steps: (i) we characterize the FIP contact-stability condition; (ii) we compute feedforward controls by solving a nonlinear optimization over receding-horizon FIP trajectories. Despite running at 30 Hz in a model-predictive fashion, simulations show that the latter is too slow to stabilize dynamic motions. To remedy this, we (iii) linearize FIP feedback control into a constrained linear-quadratic regulator that runs at 300 Hz. We finally demonstrate our solution in simulations with a model of the HRP-4 humanoid robot, including noise and delays over state estimation and foot force control.

As small legged robots for search and rescue are deployed in collapsed buildings with unstructured terrain and steep drop-offs, they run the risk of flipping over and becoming incapacitated. A single degree of freedom low-mass tail is added to a 77.5 g, 18 cm long VelociRoACH legged robot with protective shell, which gives it the capability to dynamically self-right. Quasi-static analysis of terrestrial self-righting gives design requirements for the tail actuator. Dynamic simulation predicts that terrestrial self-righting is slower than aerial self-righting with a massive tail, but can be achieved without adding significant mass away from the base of leg support. Open-loop experiments on terrain with varying friction and roughness show that VelociRoACH can dynamically self-right using tail contact in as little time as 256 ms. Finally, an autonomous self-righting experiment on an obstacle with multiple step drops demonstrates that the robot can detect inversion and rapidly self-right while walking on challenging terrain.

This paper presents a framework which allows a quadrupedal robot to execute dynamic gaits including trot, pace and dynamic lateral walk, as well as a smooth transition between them. Our method relies on an online ZMP based motion planner which continuously updates the reference motion trajectory as a function of the contact schedule and the state of the robot. The planner is coupled with a hierarchical whole-body controller which optimizes the whole-body motion and contact forces by solving a cascade of prioritized tasks. We tested our framework on ANYmal, a fully torque controllable quadrupedal robot which is actuated by series-elastic actuators.

In this paper, dual-task performance assessment robot (DAR) using projection is developed. Falling is a common problem in the growing elderly population. Fall-risk assessment systems have proven to be helpful in community-based fall prevention programs. One of the risk factors of falling is the deterioration of a person's dual-task performance. For example, gait training, which enhances both motor and cognitive functions, is a multi-target stepping task (MTST), in which participants step on assigned colored targets. To evaluate the dual-task performance during MTST in human living space, projection mapping and robot navigation to maintain a safe distance from the participant are key technologies. Projection mapping is used to evaluate the long-distance dual-task performance, where MTST images are displayed on the floor by the moving DAR. To evaluate the accuracy of the projected target position, experiments for MTST projection using the moving DAR and video analysis are carried out. Additionally, to verify the validity of the MTST by the moving DAR at a constant speed, experiments with several young participants are carried out.

We propose a novel omnidirectional stereo camera setup that is formed by two ultra-wide field-of-view (FOV) fisheye cameras. The proposed configuration is formed by two 245-degree FOV fisheye cameras, facing opposite directions, that are rigidly mounted on two sides of a rod. The overlapping view in the two fisheye images forms a ring-shaped spatial stereo setup. Our system provides stereo observations with full 360-degree FOV in horizontal directions and 65-degree FOV in the vertical direction. In addition, the two fisheye cameras altogether also provide full spherical monocular coverage of the surrounding environment. We address challenges in camera modeling, fisheye intrinsic calibration, stereo self-calibration, and depth estimation. We develop a lens-specific fisheye camera calibration method that uses manufacturer data for the optical lens to assist with the intrinsic calibration and develop an online self-calibration approach for estimating stereo extrinsic parameters. The overlapping camera views are rectified into stereo image pairs, from which a spatial stereo matching pipeline is developed for depth estimation in all horizontal directions. We show both qualitative and quantitative analysis to validate our approach.

We describe a method for formation-change trajectory planning for large quadrotor teams in obstacle-rich environments. Our method decomposes the planning problem into two stages: a discrete planner operating on a graph representation of the workspace, and a continuous refinement that converts the non-smooth graph plan into a set of Ck-continuous trajectories, locally optimizing an integral-squared-derivative cost. We account for the downwash effect, allowing safe flight in dense formations. We demonstrate the computational efficiency in simulation with up to 200 robots and the physical plausibility with an experiment with 32 nano-quadrotors. Our approach can compute safe and smooth trajectories for hundreds of quadrotors in dense environments with obstacles in a few minutes.

Bridging the `reality gap' that separates simulated robotics from experiments on hardware could accelerate robotic research through improved data availability. This paper explores domain randomization, a simple technique for training models on simulated images that transfer to real images by randomizing rendering in the simulator. With enough variability in the simulator, the real world may appear to the model as just another variation. We focus on the task of object localization, which is a stepping stone to general robotic manipulation skills. We find that it is possible to train a real-world object detector that is accurate to 1.5 cm and robust to distractors and partial occlusions using only data from a simulator with non-realistic random textures. To demonstrate the capabilities of our detectors, we show they can be used to perform grasping in a cluttered environment. To our knowledge, this is the first successful transfer of a deep neural network trained only on simulated RGB images (without pre-training on real images) to the real world for the purpose of robotic control.

Position-based proportional derivative (PD) controllers are known to be able to render compliant behaviors to a robot, and they are usually used in conjunction with a friction compensator to improve control performance. Existing methods are effective when applied to actuation systems with mechanical transmissions; however, they cannot be applied to actuation systems that use fluid transmissions owing to the characteristics resulting from fluid parameters. To solve this problem, we propose a stability-guaranteed PD control method that incorporates two observers: one for observing friction and the other is for observing flexible joint effects due to fluid compliance and internal leakage. This allows robots with fluid transmissions to asymptotically converge to the desired position. We verified the proposed approach by conducting simulations and experiments.

This paper explains low-latency joint torque feedback control based on torque estimation on each joint for gear-driven humanoid robots with harmonic drives. Force control of gear-driven robots has an advantage in its fully variable stiffness in comparison with elastic robots. However, feedback latency makes gear-driven robots vulnerable to impact rising in several milliseconds. It would be resolved by low-latency torque feedback loop in a single joint, but torque sensors are too large for life-sized humanoid robots. We estimate joint torque from motor current and rotation observed in each joint, and give artificial elasticity to joints using compliance control and shock absorption control. Our controller performance is demonstrated by landing experiments.