RSS 2016 Papers

This paper introduces a new kinematic model to describe the planar motion of an Autonomous Underwater Vehicle (AUV) moving in constant current flows. The AUV is modeled as a rigid body moving at maximum attainable forward velocity with symmetric bounds on the control input for the turning rate. The model incorporates the effect a flow will induce on the turning rate of the AUV due to the non-symmetric geometry of the vehicle. The model is then used to characterize and construct the minimum time paths that take the AUV from a given initial configuration to a final configuration in the plane. Two algorithms for the time-optimal path synthesis problem are also introduced along with several simulations to validate the proposed method.

In this paper, we propose an approach to designing online feedback controllers for input-saturated robotic systems evolving on Lie groups by extending the recently developed Sequential Action Control (SAC). In contrast to existing feedback controllers, our approach poses the nonconvex constrained non- linear optimization problem as the tracking of a desired negative mode insertion gradient on the configuration space of a Lie group. This results in a closed-form feedback control law even with input saturation and thus is well suited for online application. In extending SAC to Lie groups, the associated mode insertion gradient is derived and the switching time optimization on Lie groups is studied. We demonstrate the efficacy and scalability of our approach in the 2D kinematic car on SE(2) and the 3D quadrotor on SE(3). We also implement iLQG on a quadrator model and compare to SAC, demonstrating that SAC is both faster to compute and has a larger basin of attraction.

Roadmaps constructed by many sampling-based motion planners coincide, in the absence of obstacles, with standard models of random geometric graphs (RGGs). Those models have been studied for several decades and by now a rich body of literature exists analyzing various properties and types of RGGs. In their seminal work on optimal motion planning Karaman and Frazzoli [31] conjectured that a sampling-based planner has a certain property if the underlying RGG has this property as well. In this paper we settle this conjecture and leverage it for the development of a general framework for the analysis of sampling-based planners. Our framework, which we call localization-tessellation, allows for easy transfer of arguments on RGGs from the free unit-hypercube to spaces punctured by obstacles, which are geometrically and topologically much more complex. We demonstrate its power by providing alternative and (arguably) simple proofs for probabilistic completeness and asymptotic (near-)optimality of probabilistic roadmaps (PRMs). Furthermore, we introduce two variants of PRMs, analyze them using our framework, and discuss the implications of the analysis.

We explore the capabilities of Auto-Encoders to fuse the information available from cameras and depth sensors, and to reconstruct missing data, for scene understanding tasks. In particular we consider three input modalities: RGB images; depth images; and semantic label information. We seek to generate complete scene segmentations and depth maps, given images and partial and/or noisy depth and semantic data. We formulate this objective of reconstructing one or more types of scene data using a Multi-modal stacked Auto-Encoder. We show that suitably designed Multi-modal Auto-Encoders can solve the depth estimation and the semantic segmentation problems simultaneously, in the partial or even complete absence of some of the input modalities. We demonstrate our method using the outdoor dataset KITTI that includes LIDAR and stereo cameras. Our results show that as a means to estimate depth from a single image, our method is comparable to the state-of-the-art, and can run in real time (i.e., less than 40ms per frame). But we also show that our method has a significant advantage over other methods in that it can seamlessly use additional data that may be available, such as a sparse point-cloud and/or incomplete coarse semantic labels.

With the increased use of high degree-of-freedom robots that must perform tasks in real-time, there is a need for fast algorithms for motion planning. In this work, we view motion planning from a probabilistic perspective. We consider smooth continuous-time trajectories as samples from a Gaussian process (GP) and formulate the planning problem as probabilistic inference. We use factor graphs and numerical optimization to perform inference quickly, and we show how GP interpolation can further increase the speed of the algorithm. Our framework also allows us to incrementally update the solution of the planning problem to contend with changing conditions. We benchmark our algorithm against several recent trajectory optimization algorithms on planning problems in multiple environments. Our evaluation reveals that our approach is several times faster than previous algorithms while retaining robustness. Finally, we demonstrate the incremental version of our algorithm on replanning problems, and show that it often can find successful solutions in a fraction of the time required to replan from scratch.

In this work, we use first principles of kinematics to provide a fundamental insight into mechanical power distribution within multi-actuator machines. Individual actuator powers—not their net sum—determine the efficiency and actuator size of a multi-joint machine. Net power delivered to the environment naturally discards important information about how that power is generated. For example, simultaneous positive and negative powers will cancel within a mechanism, wasting energy and raising peak power requirements. The same effect can bias power draw toward a single actuator while the other actuators do zero work. In general, it is best for all actuators to contribute equally to the net power demand because balance minimizes the mechanical power requirements of individual actuators. In this paper, we present the actuation power space, within which we measure the antagonism in a machine (joints working against each other due to kinematic constraints). We show the difference between the net power consumed by the task and the total power supplied by the actuators. We derive the power quality measure as a smooth objective function which encodes both antagonism and the balance of power between actuators. As a demonstration of our general framework, we apply our technique to a legged-robot design to find improved kinematics for performing a running gait. This technique finds mechanisms with optimal power distribution, regardless of actuator choice or loss models, so it can be applied early in the design process using mechanism kinematics alone. After choosing appropriate kinematics for an application, designers can independently optimize each actuator in a design to minimize local losses.

A set of practical mathematical tools for creating self correcting, self compensating optical angular encoders is presented. Included is [a] a discussion and proof of the so-called Angular Encoder Theorem, which is the fundamental element of real time self correction based on read head symmetry aver- aging; [b] a discussion and derivation of the self-compensation equations, which can be used in tandem with the encoder theorem to increase accuracy further; [c] a kinematic model which can be used to simulate realistic operational scenarios; and [d] rules of thumb and implications for the designer. Sample simulation results and sample test results are also presented. These tools enable the designer to work out cost-effective encoder designs to meet different accuracy requirements for different applications. Such encoders automatically retain their accuracy over indefinite periods of time without human intervention of any kind and are thus well suited for challenging robotic applications. These tools can and have been used to construct cost-effective encoders with demonstrated sub-arcsecond accuracy. Finally, sample source code is available online.

We describe the winning entry to the Amazon Picking Challenge. From the experience of building this system and competing in the Amazon Picking Challenge, we derive several conclusions: 1) We suggest to characterize robotic systems building along four key aspects, each of them spanning a spectrum of solutions: modularity vs. integration, generality vs. assumptions, computation vs. embodiment, and planning vs. feedback. 2) To understand which region of each spectrum most adequately addresses which robotic problem, we must explore the full spectrum of possible approaches. To achieve this, our community should agree on key aspects that characterize the solution space of robotic systems. 3) For manipulation problems in unstructured environments, certain regions of each spectrum match the problem most adequately, and should be exploited further. This is supported by the fact that our solution deviated from the majority of the other challenge entries along each of the spectra.

We describe a combined force and distance sensor using a commodity infrared distance sensor embedded in a transparent elastomer with applications in robotic manipulation. Prior to contact, the sensor works as a distance sensor (0–10 cm), whereas after contact the material doubles as a spring, with force proportional to the compression of the elastomer (0– 5 N). We describe its principle of operation and de- sign parameters, including polymer thickness, mixing ratio, and emitter current, and show that the sensor response has an inflection point at contact that is independent of an object’s surface properties. We then demonstrate how two arrays of eight sensors, each mounted on a standard Baxter gripper, can be used to (1) improve gripper alignment during grasping, (2) determine contact points with objects, and (3) obtain crude 3D models that can serve to determine possible grasp locations.

We present a new algorithm for task and motion planning (TMP) and discuss the requirements and abstrac- tions necessary to obtain robust solutions for TMP in general. Our Iteratively Deepened Task and Motion Planning (IDTMP) method is probabilistically-complete and offers improved per- formance and generality compared to a similar, state-of-the- art, probabilistically-complete planner. The key idea of IDTMP is to leverage incremental constraint solving to efficiently add and remove constraints on motion feasibility at the task level. We validate IDTMP on a physical manipulator and evaluate scalability on scenarios with many objects and long plans, showing order-of-magnitude gains compared to the benchmark planner and a four-times self-comparison speedup from our extensions. Finally, in addition to describing a new method for TMP and its implementation on a physical robot, we also put forward requirements and abstractions for the development of similar planners in the future.

Snake robots can contact their environments along their whole bodies. This distributed contact makes them versatile and robust locomotors, but also makes controlling them a chal- lenging problem involving high-dimensional configuration spaces, with no direct way to break their motion down into “driving” and “steering” actions. In this paper, we use concepts from geometric mechanics—e.g., expanded Lie bracket analysis—to simplify the problem of controlling a snake robot moving across a granular surface. Without needing force laws that model the interaction of the snake robot with the granular surface, the relationship between shape and body velocities can be experimentally derived by perturbing the robot’s shape from a sampling of initial configurations, which allows us to: 1. Generate an intuitive and visualizable relationship between gait cycles and the motion they induce; 2. Make accurate predictions about the most efficient gaits available to the robot; and 3. Identify an effective turning gait for the robot that to the best of our knowledge has not previously appeared in the snake robot literature. This geometric analysis of snake robot locomotion serves as a demonstration of how differential-geometric tools can provide insight into the motion of systems that do not have the analytic models often associated with such approaches.

We introduce a functional gradient descent tra- jectory optimization algorithm for robot motion planning in Reproducing Kernel Hilbert Spaces (RKHSs). Functional gra- dient algorithms are a popular choice for motion planning in complex many-degree-of-freedom robots, since they (in theory) work by directly optimizing within a space of continuous trajec- tories to avoid obstacles while maintaining geometric properties such as smoothness. However, in practice, implementations such as CHOMP and TrajOpt typically commit to a fixed, finite parametrization of trajectories, often as a sequence of waypoints. Such a parameterization can lose much of the benefit of reasoning in a continuous trajectory space: e.g., it can require taking an inconveniently small step size and large number of iterations to maintain smoothness. Our work generalizes functional gradient trajectory optimization by formulating it as minimization of a cost functional in an RKHS. This generalization lets us represent trajectories as linear combinations of kernel functions. As a re- sult, we are able to take larger steps and achieve a locally optimal trajectory in just a few iterations. Depending on the selection of kernel, we can directly optimize in spaces of trajectories that are inherently smooth in velocity, jerk, curvature, etc., and that have a low-dimensional, adaptively chosen parameterization. Our experiments illustrate the effectiveness of the planner for different kernels, including Gaussian RBFs with independent and coupled interactions among robot joints, Laplacian RBFs, and B-splines, as compared to the standard discretized waypoint representation.

Many applications in robotics such as registration, object tracking, sensor calibration, etc. use Kalman filters to estimate a time invariant SE(3) element by locally linearizing a non-linear measurement model. Linearization-based filters tend to suffer from inaccurate estimates, and in some cases divergence, in the presence of large initialization errors. In this work, we use a dual quaternion to represent the S E (3) element and use multiple measurements simultaneously to rewrite the measurement model in a truly linear form with state dependent measurement noise. Use of the linear measurement model bypasses the need for any linearization in prescribing the Kalman filter, resulting in accurate estimates while being less sensitive to initial estimation error. To show the broad applicability of this approach, we derive linear measurement models for applications that use either position measurements or pose measurements. A procedure to estimate the state dependent measurement uncertainty is also discussed. The efficacy of the formulation is illustrated using simulations and hardware experiments for two applications in robotics: rigid registration and sensor calibration.

Our goal is to develop models that allow a robot to understand natural language instructions in the context of its world representation. Contemporary models learn possible correspondences between parsed instructions and candidate groundings that include objects, regions and motion constraints. However, these models cannot reason about abstract concepts expressed in an instruction like, “pick up the middle block in the row of five blocks”. In this work, we introduce a probabilistic model that incorporates an expressive space of abstract spatial concepts as well as notions of cardinality and ordinality. The graph is structured according to the parse structure of language and introduces a factorisation over abstract concepts correlated with concrete constituents. Inference in the model is posed as an approximate search procedure that leverages partitioning of the joint in terms of concrete and abstract factors. The algorithm first estimates a set of probable concrete constituents that constrains the search procedure to a reduced space of abstract concepts, pruning away improbable portions of the exponentially- large search space. Empirical evaluation demonstrates accurate grounding of abstract concepts embedded in complex natural language instructions commanding a robot manipulator. The proposed inference method leads to significant efficiency gains compared to the baseline, with minimal trade-off in accuracy

We introduce a novel paradigm for model-based multi-object recognition and 3 DoF pose estimation from 3D sensor data that integrates exhaustive global reasoning with discriminatively-trained algorithms in a principled fashion. Typ- ical approaches for this task are based on scene-to-model feature matching or regression by statistical learners trained on a large database of annotated scenes. These approaches are fast but sensitive to occlusions, features, and/or training data. Generative approaches, on the other hand, e.g., methods based on rendering and verification, are robust to occlusions and require no training, but are slow at test time. We conjecture that robust and efficient perception can be achieved through a combination of generative methods and discriminatively-trained approaches. To this end, we introduce the Discriminatively-guided Deliberative Perception (D2P) paradigm that has the following desirable properties: a) D2P is a single search algorithm that looks for the ‘best’ rendering of the scene that matches the input, b) can be guided by any and multiple discriminative algorithms, and c) generates a solution that is provably bounded suboptimal with respect to the chosen cost function. In addition, we introduce the notions of completeness and resolution completeness for multi-object pose estimation problems, and show that D2P is resolution complete. We conduct extensive evaluations on a benchmark dataset to study various aspects of D2P in relation to existing approaches.

Although methods about building fish-like robots have attracted much research in recent years, new techniques are still required for developing superior robot fish. This paper presents a novel robot fish propelled by an active compliant propulsion mechanism. The key innovation is the combination of an active wire-driven mechanism with a soft compliant tail to accomplish subcarangiform swimming (i.e., swimming with an “S” motion). Additionally, an effective design methodology based on a 3R pseudo-rigid-body model is proposed to design the compliant tail. First, the mathematical model based on the design methodology was derived and computer simulations were performed. Second, a prototype was fabricated and numerous experimental studies were conducted. The experiments demonstrated that the predictions of the mathematical model matched the testing results. Compared with existing robot fishes which use multi-link structure with each joint being actuated by one motor, the new robot fish has several advantages: it is simple in structure, easy to control, and capable of high speed swimming and maneuverable swimming, the maximum swimming speed reached 2.15 body length per second and the instantaneous maximum turning speed is 269°/s. Furthermore, the design methodology presented in the paper can be also used in other applications such as flexible probe of medical devices and soft manipulators.

Coordinated control strategies for multi-robot sys- tems are necessary for tasks that cannot be executed by a single robot. This encompasses tasks where the workspace of the robot is too small or where the load is too heavy for one robot to handle. Using multiple robots makes the task feasible by extending the workspace and/or increase the payload of the overall robotic system. In this paper, we consider two instances of such tasks: a co-worker scenario in which a human hands over a large object to a robot; intercepting a large flying object. The problem is made difficult as the pick-up/intercept motions must take place while the object is in motion and because the object’s motion is not deterministic. The challenge is then to adapt the motion of the robotic arms in coordination with one another and with the object. Determining the pick-up/intercept point is done by taking into account the workspace of the multi-arm system. The point is continuously recomputed to adapt to change in the object’s trajectory. We propose a virtual object based dynamical systems (DS) control law to generate autonomous and synchronized motions for a multi-arm robot system. We show theoretically that the multi-arm + virtual object system converges asymptotically to the moving object. We validate our approach on a dual-arm robotic system and demonstrate that it can re- synchronize and adapt the motion of each arm in a fraction of a second, even when the motion of the object is fast and not accurately predictable.

This work addresses the problem of how a robot can improve a manipulation skill in a sample-efficient and secure manner. As an alternative to the standard reinforcement learning formulation where all objectives are defined in a single reward function, we propose a generalized formulation that consists of three components: 1) A known analytic control cost function; 2) A black-box return function; and 3) A black-box binary success constraint. While the overall policy optimization problem is high- dimensional, in typical robot manipulation problems we can assume that the black-box return and constraint only depend on a lower-dimensional projection of the solution. With our formulation we can exploit this structure for a sample-efficient learning framework that iteratively improves the policy with respect to the objective functions under the success constraint. We employ efficient 2nd-order optimization methods to optimize the high-dimensional policy w.r.t. the analytic cost function while keeping the lower dimensional projection fixed. This is alternated with safe Bayesian optimization over the lower-dimensional projection to address the black-box return and success constraint. During both improvement steps the success constraint is used to keep the optimization in a secure region and to clearly distinguish between motions that lead to success or failure. The learning algorithm is evaluated on a simulated benchmark problem and a door opening task with a PR2.

Magnetic field gradients have repeatedly been shown to be the most feasible mechanism for gastrointestinal capsule endoscope actuation. An inverse quartic magnetic force varia- tion with distance results in large force gradients induced by small movements of a driving magnet; this necessitates robotic actuation of magnets to implement stable control of the device. A typical system consists of a serial robot with a permanent magnet at its end effector that actuates a capsule with an embedded permanent magnet. We present a tethered capsule system where a capsule with an embedded magnet is closed loop controlled in 2 degree-of-freedom in position and 2 degree-of- freedom in orientation. Capitalizing on the magnetic field of the external driving permanent magnet, the capsule is localized in 6- D allowing for both position and orientation feedback to be used in a control scheme. We developed a relationship between the serial robot’s joint parameters and the magnetic force and torque that is exerted onto the capsule. Our methodology was validated both in a dynamic simulation environment where a custom plug-in for magnetic interaction was written, as well as on an experimental platform. The tethered capsule was demonstrated to follow desired trajectories in both position and orientation with accuracy that is acceptable for colonoscopy.