ICRA 2018 Papers

We propose a simple drop-in noise-tolerant replacement for the standard finite difference procedure used ubiquitously in blackbox optimization. In our approach, parameter perturbation directions are defined by a family of structured orthogonal matrices. We show that at the small cost of computing a Fast Walsh-Hadamard/Fourier Transform (FWHT/FFT), such structured finite differences consistently give higher quality approximation of gradients and Jacobians in comparison to vanilla approaches that use coordinate directions or random Gaussian perturbations. We find that trajectory optimizers like Iterative LQR and Differential Dynamic Programming require fewer iterations to solve several classic continuous control tasks when our methods are used to linearize noisy, blackbox dynamics instead of standard finite differences. By embedding structured exploration in a quasi-Newton optimizer (LBFGS), we are able to learn agile walking and turning policies for quadruped locomotion, that successfully transfer from simulation to actual hardware. We theoretically justify our methods via bounds on the quality of gradient reconstruction and provide a basis for applying them also to nonsmooth problems.

In this paper, we propose a novel solution to optimize the deployment of Radio Frequency (RF) beacons for the purpose of indoor localization. We propose a system that optimizes both the number of beacons and their placement in a given environment. We propose a novel cost-function, called CovBsm, that allows to simultaneously optimize the 3-coverage while maximizing the beacon spreading. Using this cost function, we propose a framework that maximize both the number of beacons and their placement in a given environment. The proposed solution accounts for the indoor infrastructure and its influence on the RF signal propagation by embedding a realistic simulator into the optimization process.

Achieving full autonomous robotic environment exploration in the underwater domain is very challenging, mainly due to noisy acoustic sensors, high localization error, control disturbances of the water and lack of accurate underwater maps. In this work we present a robotic exploration algorithm for underwater vehicles that does not rely on prior information about the environment. Our method has been greatly influenced by many robotic exploration, view planning and path planning algorithms. The proposed method constitutes a significant improvement over our previous work [1]: Firstly, we refine our exploration approach to improve robustness; Secondly, we propose an alternative map representation based on the quadtree data structure that allows different relevant queries to be performed efficiently, reducing the computational cost of the viewpoint generation process; Thirdly, we present an algorithm that is capable of generating consistent maps even when noisy sonar data is used. The aforementioned contributions have increased the reliability of the algorithm, allowing new real experiments performed in artificial structures but also in more challenging natural environments, from which we provide a 3D reconstruction to show that with this algorithm full optical coverage is obtained.

To operate intelligently in domestic environments, robots require the ability to understand arbitrary spatial relations between objects and to generalize them to objects of varying sizes and shapes. In this work, we present a novel end-to-end approach to generalize spatial relations based on distance metric learning. We train a neural network to transform 3D point clouds of objects to a metric space that captures the similarity of the depicted spatial relations, using only geometric models of the objects. Our approach employs gradient-based optimization to compute object poses in order to imitate an arbitrary target relation by reducing the distance to it under the learned metric. Our results based on simulated and real-world experiments show that the proposed method enables robots to generalize spatial relations to unknown objects over a continuous spectrum.

Vibratory bowl feeders (VBFs) are a widely used option for industrial part feeding, but their design is still largely manual. A subtask of VBF design is determining an optimal parameter set for the passive devices, called traps, which the VBF uses to ensure correct part orientation. This paper proposes a fast and robust strategy for optimising traps, which makes use of dynamic simulation to efficiently evaluate the performance of parameter sets. The optimisation strategy is based on Bayesian Optimisation and selects new parameter sets to evaluate, using a modified Upper Confidence Bound with regression by Kernel Density Estimation as function estimator. The optimisation is run for four different traps with an industrial part and the best parameter sets are tested for robustness in simulation. The traps are then combined to create two sequences performing orientation of the parts and the designs are prototyped and tested on a real VBF.

Uncertainties in flow models have to be explicitly considered for effective path planning in marine environments. In this paper, we present two methods to compute minimum expected cost policies and paths over an uncertain flow model. The first method based on a Markov Decision Process computes a minimum expected cost policy while the second graph search based method, computes a minimum expected cost path. A transition probability model is developed to compute the probability of transition from one state to another under a given action. In addition, a method to compute the expected cost of a path when it is executed in an uncertain flow field is also presented. The two methods are used to compute minimum energy paths in an ocean environment and the results are analyzed in simulations.

In this paper, we formulate an environmental sensing problem for multi-robot teams that couples intermittent deployments with the selection of team composition and sensor type over time. We suppose that a multi-robot team needs to autonomously sense an environmental process and find the optimal policy for deploying heterogeneous robots. In addition, heterogeneous robot teams can be composed in various ways by selecting different mobility and sensor types which have varying accuracies and costs, resulting in a more complex problem. The question is then how to find an optimal intermittent deployment and sensor selection policy that captures both cost and estimation accuracy based on partial environmental information. By utilizing structural results from partially observable Markov decision processes (POMDP) and exploiting submodularity, an optimal policy, which minimizes cost while maintaining a high accuracy, can be achieved in this paper. The effectiveness of this method is demonstrated by simulation results and comparisons with naive policies.

The goal of this paper is to increase the estimation performance of an Extended Kalman Filter for a nonlinear differentially flat system by planning trajectories able to maximize the amount of information gathered by onboard sensors in presence of both process and measurement noises. In a previous work, we presented an online gradient descent method for planning optimal trajectories along which the smallest eigenvalue of the Observability Gramian (OG) is maximized. As the smallest eigenvalue of the OG is inversely proportional to the maximum estimation uncertainty, its maximization reduces the maximum estimation uncertainty of any estimation algorithm employed during motion. However, the OG does not consider the process noise that, instead, in several applications is far from being negligible. For this reason, this paper proposes a novel solution able to cope with non-negligible process noise: this is achieved by minimizing the largest eigenvalue of the a posteriori covariance matrix obtained by solving the Continuous Riccati Equation as a measure of the total available information. This minimization is expected to maximize the information gathered by the outputs while, at the same time, limiting as much as possible the negative effects of the process noise. We apply our method to a unicycle robot. The comparison between the novel method and the one of our previous work (which did not consider process noise) shows significant improvements in the obtained estimation accuracy.

A low-cost optical sensing method for improved measurement and control of soft pneumatic manipulator motion is presented. The core of a soft continuum robot is embedded with several optically-diffuse elastomer sensors which attenuate light depending on their strain mode and degree. The optical sensors measure local strains at the robot's axial center, and these strain data are combined with measured actuator chamber pressures to determine the pose of the robot under various gravitational and tip loading conditions. Regression analyses using neural networks (NNs) demonstrate that when the soft continuum robot's base orientation is fixed, the position of its end-effector can be estimated with 3.42 times more accuracy (71 % smaller root mean squared error) when using both optical sensor and pressure data (~2.44mm) than when using only pressure data (~8.3mm). When the robot's base orientation was varied, the combined optical sensor and pressure data provide position estimates which are as much as 37.8 times more accurate (~2.76mm) than pressure data alone (~104mm).

In this work, we propose and experimentally validate a novel optical fiber-based sensor for monitoring and assessing electric current in unmanned aerial vehicles electric motors. The proposed sensing technology combines a Long-Period Fiber Grating sensor and a permanent Neodymium magnet, providing a small and flexible sensing scheme deployed inside the arm of the drone. The experimental results show that good accuracy and linear electric current sensitivity of 0.21 A and 2.08 A/nm, respectively, were achieved with electric current measurements at a 100 Hz sampling rate. The values of hysteresis and repeatability achieved were 0.08 A and 0.22 A, respectively. Finally, a Robot Operating System package for interfacing with the sensing system was developed and tested, which greatly simplifies the deployment of the sensor in robotics applications.

While deep reinforcement learning techniques have recently produced considerable achievements on many decision-making problems, their use in robotics has largely been limited to simulated worlds or restricted motions, since unconstrained trial-and-error interactions in the real world can have undesirable consequences for the robot or its environment. To overcome such limitations, we propose a novel reinforcement learning architecture, OptLayer, that takes as inputs possibly unsafe actions predicted by a neural network and outputs the closest actions that satisfy chosen constraints. While learning control policies often requires carefully crafted rewards and penalties while exploring the range of possible actions, OptLayer ensures that only safe actions are actually executed and unsafe predictions are penalized during training. We demonstrate the effectiveness of our approach on robot reaching tasks, both simulated and in the real world.

The use ofrobotic devices for drug delivery in sensitive area of the human body is an innovative and reliable solution. Most of them use magnetics fields, to steer micro-nano-robots into diseases locations. In this study, we use a magnetic actuator based on two permanent magnets as an end-effector of a robotic manipulator. The actuator offers the possibility to generate both pushing and pulling forces on the magnetic actuator axis in an open-loop control way. We describe in this paper the robotic drug delivery strategy that we implemented in a 6 degree of freedom robotic manipulator to push and to steer a magnetic microparticle from the round window to the apex of the human cochlea. Different experiments have been conducted in order to demonstrate the effectiveness and robustness of the navigation proposed strategy on a human phantom cochlea. The results demonstrate clearly that precise and reliable drug administration is rendered possible.

This paper is concerned with system identification and the calibration of parameters of dynamic models used in different robotic platforms. A constant time algorithm has been developed in order to automatically calibrate the parameters of a high-fidelity dynamical model for a robotic platform. The presented method is capable of choosing informative motion segments in order to calibrate model parameters in constant time while also calculating a confidence level on each estimated parameter. Simulations and experiments with a ⅛ th scale four wheel drive vehicle are performed to calibrate two of the parameters of test vehicle which demonstrate the accuracy and efficiency of the approach.

In this paper, we propose a framework for online quadrotor motion planning for autonomous navigation in unknown environments. Based on the onboard state estimation and environment perception, we adopt a fast marching-based path searching method to find a path on a velocity field induced by the Euclidean signed distance field (ESDF) of the map, to achieve better time allocation. We generate a flight corridor for the quadrotor to travel through by inflating the path against the environment. We represent the trajectory as piecewise Bézier curves by using Bernstein polynomial basis and formulate the trajectory generation problem as typical convex programs. By using Bézier curves, we are able to bound positions and higher order dynamics of the trajectory entirely within safe regions. The proposed motion planning method is integrated into a customized light-weight quadrotor platform and is validated by presenting fully autonomous navigation in unknown cluttered indoor and outdoor environments. We also release our code for trajectory generation as an open-source package.

Sparse feature-based maps provide a compact representation of the environment that admit efficient algorithms, for example simultaneous localization and mapping. These representations typically assume a static world and therefore contain static map features. However, since the world contains dynamic elements, determining when map features no longer correspond to the environment is essential for long-term utility. This work develops a feature-based model of the environment which evolves over time through feature persistence. Moreover, we augment the state-of-the-art sparse mapping model with a correlative structure that captures spatio-temporal properties, e.g. that nearby features frequently have similar persistence. We show that such relationships, typically addressed through an ad hoc formalism focusing only on feature repeatability, are crucial to evaluate through a probabilistically principled approach. The joint posterior over feature persistence can be computed efficiently and used to improve online data association decisions for localization. The proposed algorithms are validated in numerical simulation and using publicly available data sets.

The promise of learning to learn for robotics rests on the hope that by extracting some information about the learning process itself we can speed up subsequent similar learning tasks. Here, we introduce a computationally efficient online meta-learning algorithm that builds and optimizes a memory model of the optimal learning rate landscape from previously observed gradient behaviors. While performing task specific optimization, this memory of learning rates predicts how to scale currently observed gradients. After applying the gradient scaling our meta-learner updates its internal memory based on the observed effect its prediction had. Our meta-learner can be combined with any gradient-based optimizer, learns on the fly and can be transferred to new optimization tasks. In our evaluations we show that our meta-learning algorithm speeds up learning of MNIST classification and a variety of learning control tasks, either in batch or online learning settings.

Most of the existing monocular visual-inertial SLAM techniques assume that the camera-IMU extrinsic parameters are known, therefore these methods merely estimate the initial values of velocity, visual scale, gravity, biases of gyroscope and accelerometer in the initialization stage. However, it's usually a professional work to carefully calibrate the extrinsic parameters, and it is required to repeat this work once the mechanical configuration of the sensor suite changes slightly. To tackle this problem, we propose an online initialization method to automatically estimate the initial values and the extrinsic parameters without knowing the mechanical configuration. The biases of gyroscope and accelerometer are considered in our method, and a convergence criteria for both orientation and translation calibration is introduced to identify the convergence and to terminate the initialization procedure. In the three processes of our method, an iterative strategy is firstly introduced to iteratively estimate the gyroscope bias and the extrinsic orientation. Secondly, the scale factor, gravity, and extrinsic translation are approximately estimated without considering the accelerometer bias. Finally, these values are further optimized by a refinement algorithm in which the accelerometer bias and the gravitational magnitude are taken into account. Extensive experimental results show that our method achieves competitive accuracy compared with the state-of-the-art with less calculation.

This paper proposes a novel fall control technique based on energy concepts, which can be applied online to mitigate the impact forces incurred during the falling over of humanoids. The technique reduces the total energy using a nonlinear control tool, called energy shaping (ES), and further distributes the reduced energy over multiple contacts by means of energy distribution polygons (EDP). We also include an effective orientation control to safeguard the end-effectors in the event of ground impacts. The performance of the proposed method is numerically evaluated by dynamic simulations under the sudden falling over scenario of the humanoid robot for both lateral and sagittal falls. The effectiveness of the proposed ES and EDP concepts are verified by diverse comparative simulations with total energy, distribution, and impact forces.

This paper focuses on real-time estimation of the aerodynamic model parameters of small-scale fixed wing Unmanned Aerial Vehicles (UAVs) without the aid of wind-tunnel experiments, using exclusively flight data. The key tool of the following analysis centers around the principles of Total Least Squares estimation. Contrary to Ordinary Least Squares, this method accounts for errors in both explanatory data and variables to-be-explained. This is a highly desirable property for UAVs equipped with low-cost sensor systems. The proposed implementation combines both batch and real-time schemes, while deals efficiently with the problem of Insufficient System Excitation. Online adaptation to model changes is performed by applying a Variable Forgetting Factor to the estimation data. Finally, a Monte Carlo approach is developed for uncertainty estimation regarding compound aerodynamic variables.