RSS 2015 Papers

When making contact with an object, a robot can use a tactile sensor consisting of a heating element and a temperature sensor to recognize the object's material based on conductive heat transfer from the tactile sensor to the object. When this type of tactile sensor has time to fully reheat prior to contact and the duration of contact is long enough to achieve a thermal steady state, numerous methods have been shown to perform well. In order to enable robots to more efficiently sense their environments and take advantage of brief contact events over which they lack control, we focus on the problem of material recognition from heat transfer given varying initial conditions and short-duration contact. We present both model-based and data-driven methods. For the model-based method, we modeled the thermodynamics of the sensor in contact with a material as contact between two semi-infinite solids. For the data-driven methods, we used three machine learning algorithms (SVM+PCA, k-NN+PCA, HMMs) with time series of raw temperature measurements and temperature change estimates. When recognizing 11 materials with varying initial conditions and 3-fold cross-validation, SVM+PCA outperformed all other methods, achieving 84% accuracy with 0.5 s of contact and 98% accuracy with 1.5 s of contact.

This paper presents the Rapidly-exploring Adaptive Search and Classification (ReASC) algorithm, a sampling-based algorithm for planning the trajectories of mobile robots performing real-time target search and classification tasks in the field. The proposed algorithm incrementally builds up a tree of solutions and evaluates the utility of each solution for identifying targets in an environment. An optimistic approximation for the classification utility is used, which reduces the computational cost of evaluating trajectories and makes real-time adaptive planning feasible. The proposed algorithm is tested on an autonomous aquatic vehicle and are shown to outperform myopic methods by up to 36% in a lake monitoring scenario.

We build on previous works advocating the use of the Gravito-Inertial Wrench Cone (GIWC) as a general contact stability criterion (a ZMP for non-coplanar contacts). We show how to compute this wrench cone from the friction cones of contact forces by using an intermediate representation, the surface contact wrench cone, which is the minimal representation of contact stability for each surface contact. The observation that the GIWC needs to be computed only once per stance leads to particularly efficient algorithms, as we illustrate in two important problems for humanoids: testing robust static equilibrium and time-optimal path parameterization. We show, through theoretical analysis and in physical simulations, that our method is more general and/or outperforms existing ones.

We propose a layered street view model to encode both depth and semantic information on street view images for autonomous driving. Recently, stixels, stix-mantics, and tiered scene labeling methods have been proposed to model street view images. We propose a 4-layer street view model, a compact representation over the recently proposed stix-mantics model. Our layers encode semantic classes like ground, pedestrians, vehicles, buildings, and sky in addition to the depths. The only input to our algorithm is a pair of stereo images. We use a deep neural network to extract the appearance features for semantic classes. We use a simple and an efficient inference algorithm to jointly estimate both semantic classes and layered depth values. Our method outperforms other competing approaches in Daimler urban scene segmentation dataset. Our algorithm is massively parallelizable, allowing a GPU implementation with a processing speed about 9 fps.

We propose an information-theoretic planning approach that enables mobile robots to autonomously construct dense 3D maps in a computationally efficient manner. Inspired by prior work, we accomplish this task by formulating an information-theoretic objective function based on Cauchy-Schwarz quadratic mutual information (CSQMI) that guides robots to obtain measurements in uncertain regions of the map. We then contribute a two stage approach for active mapping. First, we generate a candidate set of trajectories using a combination of global planning and generation of local motion primitives. From this set, we choose a trajectory that maximizes the information-theoretic objective. Second, we employ a gradient-based trajectory optimization technique to locally refine the chosen trajectory such that the CSQMI objective is maximized while satisfying the robot's motion constraints. We evaluated our approach through a series of simulations and experiments on a ground robot and an aerial robot mapping unknown 3D environments. Real-world experiments suggest our approach reduces the time to explore an environment by 70% compared to a closest frontier exploration strategy and 57% compared to an information-based strategy that uses global planning, while simulations demonstrate the approach extends to aerial robots with higher-dimensional state.

Recent results in monocular visual-inertial navigation (VIN) have shown that optimization-based approaches outperform filtering methods in terms of accuracy due to their capability to relinearize past states. However, the improvement comes at the cost of increased computational complexity. In this paper, we address this issue by preintegrating inertial measurements between selected keyframes. The preintegration allows us to accurately summarize hundreds of inertial measurements into a single relative motion constraint. Our first contribution is a preintegration theory that properly addresses the manifold structure of the rotation group and carefully deals with uncertainty propagation. The measurements are integrated in a local frame, which eliminates the need to repeat the integration when the linearization point changes while leaving the opportunity for belated bias corrections. The second contribution is to show that the preintegrated IMU model can be seamlessly integrated in a visual-inertial pipeline under the unifying framework of factor graphs. This enables the use of a structureless model for visual measurements, further accelerating the computation. The third contribution is an extensive evaluation of our monocular VIN pipeline: experimental results confirm that our system is very fast and demonstrates superior accuracy with respect to competitive state-of-the-art filtering and optimization algorithms, including off-the-shelf systems such as Google Tango.

The vast amount of data robots can capture today motivates the development of fast and scalable statistical tools to model the environment the robot operates in. We devise a new technique for environment representation through continuous occupancy mapping that improves on the popular occupancy grip maps in two fundamental aspects: 1) it does not assume an a priori discretisation of the world into grid cells and therefore can provide maps at an arbitrary resolution; 2) it captures statistical relationships between measurements naturally, thus being more robust to outliers and possessing better generalisation performance. The technique named Hilbert maps, is based on the computation of fast kernel approximations that project the data in a Hilbert space where a logistic regression classifier is learnt. We show that this approach allows for efficient stochastic gradient descent optimisation where each measurement is only processed once during learning or online updates. We present results with three types of kernel approximations, Random Fourier, Nystrom and a novel sparse projections. We also show how to extend the approach to accept probability distributions as inputs, i.e. when there is uncertainty over the position of laser scans due to sensor or localisation errors. Experiments demonstrate the benefits of the approach in popular benchmark datasets with several thousand laser scans.

Unmanned aerial vehicle (UAV) capability is currently limited by the amount of energy that can be stored onboard. Airborne docking, for mid-air refueling, is a viable solution that has been implemented with manned aircraft for decades, but has yet to be achieved with their unmanned counterparts. The prohibitive challenge is the highly accurate and reliable relative positioning performance that is necessary to dock with a small target, in the air, amidst external disturbances. This paper presents a complete solution to airborne docking, which includes vision-aided unscented Kalman filters for leader-relative navigation and docking appendage motion estimation; and guidance that is suitable for all phases of the mission. The work concludes by demonstrating the proposed algorithms in what is thought to be the first UAV airborne docking.

Multi-robot networks use wireless communication to provide wide-ranging services such as aerial surveillance and unmanned delivery. However, effective coordination between multiple robots requires trust, making them particularly vulnerable to cyber-attacks. Specifically, such networks can be gravely disrupted by the Sybil attack, where even a single malicious robot can spoof a large number of fake clients. This paper proposes a new solution to defend against the Sybil attack, without requiring expensive cryptographic key-distribution. Our core contribution is a novel algorithm implemented on commercial Wi-Fi radios that can "sense" spoofers using the physics of wireless signals. We derive theoretical guarantees on how this algorithm bounds the impact of the Sybil Attack on a broad class of robotic coverage problems. We experimentally validate our claims using a team of AscTec quadrotor servers and iRobot Create ground clients, and demonstrate spoofer detection rates over 96%.

As intelligent robots become more prevalent, methods to make interaction with the robots more accessible are increasingly important. Communicating the tasks that a person wants the robot to carry out via natural language, and training the robot to ground the natural language through demonstration, are especially appealing approaches for interaction, since they do not require a technical background. However, existing approaches map natural language commands to robot command languages that directly express the sequence of actions the robot should execute. This sequence is often specific to a particular situation and does not generalize to new situations. To address this problem, we present a system that grounds natural language commands into reward functions using demonstrations of different natural language commands being carried out in the environment. Because language is grounded to reward functions, rather than explicit actions that the robot can perform, commands can be high-level, carried out in novel environments autonomously, and even transferred to other robots with different action spaces. We demonstrate that our learned model can be both generalized to novel environments and transferred to a robot with a different action space than the action space used during training.

Accurately estimating a robot's pose relative to a global scene model and precisely tracking the pose in real-time is a fundamental problem for navigation and obstacle avoidance tasks. Due to the computational complexity of localization against a large map and the memory consumed by the model, state-of-the-art approaches are either limited to small workspaces or rely on a server-side system to query the global model while tracking the pose locally. The latter approaches face the problem of smoothly integrating the server's pose estimates into the trajectory computed locally to avoid temporal discontinuities. In this paper, we demonstrate that large-scale, real-time pose estimation and tracking can be performed on mobile platforms with limited resources without the use of an external server. This is achieved by employing map and descriptor compression schemes as well as efficient search algorithms from computer vision. We derive a formulation for integrating the global pose information into a local state estimator that produces much smoother trajectories than current approaches. Through detailed experiments, we evaluate each of our design choices individually and document its impact on the overall system performance, demonstrating that our approach outperforms state-of-the-art algorithms for localization at scale.

In this paper we point out an overlooked structure of SLAM that distinguishes it from a generic nonlinear least squares problem. The measurement function in most common forms of SLAM is linear with respect to robot and features positions. Therefore, given an estimate for robot orientation, the conditionally optimal estimate for the rest of state variables can be easily obtained by solving a sparse linear-Gaussian estimation problem. We propose an algorithm to exploit this intrinsic property of SLAM by stripping the problem down to its nonlinear core, while maintaining its natural sparsity. Our algorithm can be used together with any Newton-based iterative solver and is applicable to 2D/3D pose-graph and feature-based problems. Our results suggest that iteratively solving the nonlinear core of SLAM leads to a fast and reliable convergence as compared to the state-of-the-art back-ends.

We present a novel approach to real-time dense visual SLAM. Our system is capable of capturing comprehensive dense globally consistent surfel-based maps of room scale environments explored using an RGB-D camera in an incremental online fashion, without pose graph optimisation or any post-processing steps. This is accomplished by using dense frame-to-model camera tracking and windowed surfel-based fusion coupled with frequent model refinement through non-rigid surface deformations. Our approach applies local model-to-model surface loop closure optimisations as often as possible to stay close to the mode of the map distribution, while utilising global loop closure to recover from arbitrary drift and maintain global consistency.

Stochastic optimal control problems frequently arise as motion control problems in the context of robotics. Unfortunately, all existing approaches that guarantee arbitrary precision suffer from the curse of dimensionality: the computational effort invested by the algorithm grows exponentially fast with increasing dimensionality of the state space of the underlying dynamic system governing the robot. In this paper, we propose a novel algorithm that utilizes compressed representations to efficiently solve stochastic optimal control problems with arbitrary precision. The running time of the new algorithms scale linearly with increasing dimensionality of the state space! The running time also depends polynomially on the rank of the value function, a measure that quantifies the intrinsic geometric complexity of the value function, due to the geometry and physics embedded in the problem instance at hand. The new algorithms are based on the recent analysis and algorithms for tensor decomposition, generalizing matrix decomposition algorithms, e.g., the singular value decomposition, to three or more dimensions. In computational experiments, we show the computational effort of the new algorithm also scales linearly with the discretization resolution of the state space. We also demonstrate the new algorithm on a problem involving the perching of an aircraft, represented by a nonlinear non-holonomic longitudinal model with a seven-dimensional state space, the full numerical solution to which was not obtained before. In this example, we estimate that the proposed algorithm runs more than seven orders of magnitude faster, when compared to the naive value iteration.

Inverse Optimal Control (IOC) has strongly impacted the systems engineering process, enabling automated planner tuning through straightforward and intuitive demonstration. The most successful and established applications, though, have been in lower dimensional problems such as navigation planning where exact optimal planning or control is feasible. In higher dimensional systems, such as humanoid robots, research has made substantial progress toward generalizing the ideas to model free or locally optimal settings, but these systems are complicated to the point where demonstration itself can be difficult. Typically, real-world applications are restricted to at best noisy or even partial or incomplete demonstrations that prove cumbersome in existing frameworks. This work derives a very flexible method of IOC based on a form of Structured Prediction known as Direct Loss Minimization. The resulting algorithm is essentially Policy Search on a reward function that rewards similarity to demonstrated behavior (using Covariance Matrix Adaptation (CMA) in our experiments). Our framework blurs the distinction between IOC, other forms of Imitation Learning, and Reinforcement Learning, enabling us to derive simple, versatile, and practical algorithms that blend imitation and reinforcement signals into a unified framework. Our experiments analyze various aspects of its performance and demonstrate its efficacy on conveying preferences for motion shaping and combined reach and grasp quality optimization.

Designing robotic controllers for tasks with complex non-linear dynamics is extremely challenging, time-consuming, and in many cases, infeasible. This difficulty is exacerbated in tasks such as robotic food-cutting, in which dynamics might vary both with environmental properties, such as material and tool class, and with time while acting. In this work, we present DeepMPC, an online real-time model-predictive control approach designed to handle such difficult tasks. Rather than hand-design a dynamics model for the task, our approach uses a novel deep architecture and learning algorithm, learning controllers for complex tasks directly from data. We validate our method in experiments on a large-scale dataset of 1488 material cuts for 20 diverse classes, and in 450 real-world robotic experiments, demonstrating significant improvement over several other approaches.

In this paper we present a novel approach for the parameterization of the trajectory of a moving platform, which facilitates the development of real-time pose-estimation methods. The key idea of the proposed approach is the decoupling of the parameterization of the trajectory estimate from the parameterization of the error in this estimate. Specifically, we represent the trajectory estimate as usual, via a set of pose states, each associated with a sensor reading (e.g., a laser scan or an image). The novelty of our approach lies in the representation of the estimation errors, for which we employ B-splines. This decoupled formulation, which we term Decoupled Estimate-Error Parameterization (DEEP) offers two key advantages. First, the use of a pose-based representation of the trajectory allows us to represent arbitrarily complex trajectories. Second, the use of B-splines for error representation allows us to control the computational complexity of an estimator, by selecting the density of the knots of the B-spline. We empirically demonstrate that, in the problem of visual-inertial localization, the DEEP formulation leads to substantial computational gains, while incurring only a small loss of estimation performance.

Rearranging multiple objects is a critical skill for robots so that they can effectively deal with clutter in human spaces. This is a challenging problem as it involves combinatorially large, continuous C-spaces involving multiple movable bodies and complex kinematic constraints. This work initially revisits an existing search-based approach, which solves monotone challenges, i.e., when objects need to be grasped only once so as to be rearranged. The first contribution is the extension of this technique to a method that addresses many non-monotone challenges. The second contribution is the use of either the monotone or of the new non-monotone method as a local planner in the context of a higher-level task planner that searches the space of object placements and which provides stronger guarantees. The paper aims to emphasize the benefit of using more powerful motion primitives in the context of task planning for object rearrangement than an individual pick-and-place. Experiments in simulation using a model of a Baxter robot arm show the capability of solving difficult instances of rearrangement problems and evaluate the methods in terms of success ratio, running time, scalability and path quality.

In this work, we present an approach to topological motion planning which is fully data-driven in nature and which relies solely on the knowledge of samples in the free configuration space. For this purpose, we discuss the use of persistent cohomology with coefficients in a finite field to compute a basis which allows us to efficiently solve the path planning problem. The proposed approach can be used both in the case where a part of a configuration space is well-approximated by samples and, more generally, with arbitrary filtrations arising from real-world data sets. Furthermore, our approach can generate motions in a subset of the configuration space specified by the sub- or superlevel set of a filtration function such as a cost function or probability distribution. Our experiments show that our approach is highly scalable in low dimensions and we present results on simulated PR2 arm motions as well as GPS trace and motion capture data.