Hannes Sommer

Quadrupedal locomotion using hierarchical operational space control

This paper presents the application of operational space control based on hierarchical task optimization for quadrupedal locomotion. We show how the behavior of a complex robotic machine can be described by a simple set of least squares problems with different priorities for motion, torque, and force optimization. Using projected dynamics of floating base systems with multiple contact points, the optimization dimensionality can be reduced or decoupled such that the formulation is purely based on the inversion of kinematic system properties. The present controller is extensively tested in various experiments using the fully torque controllable quadrupedal robot StarlETH. The load distribution is optimized for static walking gaits to improve contact stability and/or actuator efficiency under various terrain conditions. This is augmented with simultaneous joint position and torque limitations as well as with an interpolation method to ensure smooth contact transitions. The same control structure is further used to stabilize dynamic trotting gaits under significant external disturbances such as uneven ground or pushes. To the best of our knowledge, this work is the first documentation of static and dynamic locomotion with pure task-space inverse dynamics (no joint position feedback) control.

Predicting actions to act predictably: Cooperative partial motion planning with maximum entropy models

This paper reports on a data-driven motion planning approach for interaction-aware, socially-compliant robot navigation among human agents. Autonomous mobile robots navigating in workspaces shared with human agents require motion planning techniques providing seamless integration and smooth navigation in such. Smooth integration in mixed scenarios calls for two abilities of the robot: predicting actions of others and acting predictably for them. The former requirement requests trainable models of agent behaviors in order to accurately forecast their actions in the future, taking into account their reaction on the robots decisions. A human-like navigation style of the robot facilitates other agents-most likely not aware of the underlying planning technique applied-to predict the robot motion vice versa, resulting in smoother joint navigation. The approach presented in this paper is based on a feature-based maximum entropy model and is able to guide a robot in an unstructured, real-world environment. The model is trained to predict joint behavior of heterogeneous groups of agents from onboard data of a mobile platform. We evaluate the benefit of interaction-aware motion planning in a realistic public setting with a total distance traveled of over 4 km. Interestingly the motion models learned from human-human interaction did not hold for robot-human interaction, due to the high attention and interest of pedestrians in testing basic braking functionality of the robot.

Automatic Differentiation on Differentiable Manifolds as a Tool for Robotics

Automatic differentiation (AD) is a useful tool for computing Jacobians of functions needed in estimation and control algorithms. However, for many interesting problems in robotics, state variables live on a differentiable manifold. The most common example are robot orientations that are elements of the Lie group SO(3). This causes problems for AD algorithms that only consider differentiation at the scalar level. Jacobians produced by scalar AD are correct, but scalar-focused methods are unable to apply simplifications based on the structure of the specific manifold. In this paper we extend the theory of AD to encompass handling of differentiable manifolds and provide a C++ library that exploits strong typing and expression templates for fast, easy-to-use Jacobian evaluation. This method has a number of benefits over scalar AD. First, it allows the exploitation of algebraic simplifications that make Jacobian evaluations more efficient than their scalar counterparts. Second, strong typing reduces the likelihood of programming errors arising from misinterpretation that are possible when using simple arrays of scalars. To the best of our knowledge, this is the first work to consider the structure of differentiable manifolds directly in AD.

ROCK∗ — Efficient black-box optimization for policy learning

Robotic learning on real hardware requires an efficient algorithm which minimizes the number of trials needed to learn an optimal policy. Prolonged use of hardware causes wear and tear on the system and demands more attention from an operator. To this end, we present a novel black-box optimization algorithm, Reward Optimization with Compact Kernels and fast natural gradient regression (ROCK*). Our algorithm immediately updates knowledge after a single trial and is able to extrapolate in a controlled manner. These features make fast and safe learning on real hardware possible. We have evaluated our algorithm on two simulated reaching tasks of a 50 degree-of-freedom robot arm and on a hopping task of a real articulated legged system. ROCK* outperformed current state-of-the-art algorithms in all tasks by a factor of three or more.

Non-uniform sampling strategies for continuous correction based trajectory estimation

Sliding window estimation is widely used for online simultaneous localization and mapping. While increasing the sliding window size generally yields improved accuracy, it also comes at an increase in computational cost. In order to reduce this cost, we propose smarter non-uniform sampling of the trajectory representation over the sliding window. This non-uniform temporal resolution is possible with continuous-time representations that allow freely adjustable knots location. Four strategies for selecting the knots location are presented and evaluated based on a real data laser-odometry SLAM problem. The results clearly show that non-uniform distributions of knots can be superior to uniform distribution in terms of accuracy per computation time.

Online self-calibration for robotic systems

We present a generic algorithm for self-calibration of robotic systems that utilizes two key innovations. First, it uses an information-theoretic measure to automatically identify and store novel measurement sequences. This keeps the computation tractable by discarding redundant information and allows the system to build a sparse but complete calibration dataset from data collected at different times. Second, as the full observability of the calibration parameters may not be guaranteed for an arbitrary measurement sequence, the algorithm detects and locks unobservable directions in parameter space using a combination of rank-revealing QR and singular value decompositions of the Fisher information matrix. The result is an algorithm that listens to an incoming sensor stream, builds a minimal set of data for estimating the calibration parameters, and updates parameters as they become observable, leaving the others locked at their initial guess. We validate our approach through an extensive set of simulated and real-world experiments.

An online multi-robot SLAM system for 3D LiDARs

Using multiple cooperative robots is advantageous for time critical Search and Rescue (SaR) missions as they permit rapid exploration of the environment and provide higher redundancy than using a single robot. A considerable number of applications such as autonomous driving and disaster response could benefit from merging mapping data from several agents. Online multi-robot localization and mapping has mainly been addressed for robots equipped with cameras or 2D LiDARs. However, in unstructured and ill-lighted real-life scenarios, a mapping system can potentially benefit from a rich 3D geometric solution. In this work, we present an online localization and mapping system for multiple robots equipped with 3D LiDARs. This system is based on incremental sparse pose-graph optimization using sequential and place recognition constraints, the latter being identified using a 3D segment matching approach. The result is a unified representation of the world and relative robot trajectories. The complete system runs in real-time and is evaluated with two experiments in different environments: one urban and one disaster scenario. The system is available open source and easy-to-run demonstrations are publicly available.

Continuous-Time Estimation of Attitude Using B-Splines on Lie Groups

Filtering algorithms are the workhorse of spacecraft attitude estimation, but recent research has shown that the use of batch estimation techniques can result in higher accuracy per unit of computational cost. This paper presents an approach for singularity-free batch estimation of attitude in continuous time using B-Spline curves on unit-length quaternions. It extends existing theory of unit-length quaternion B-splines to general Lie groups and arbitrary B-spline order. It is shown how to use these curves for continuous-time batch estimation using Gauss–Newton or Levenberg–Marquardt, including efficient curve initialization, a parameter update step that preserves the Lie group constraint within an unconstrained optimization framework, and the derivation of Jacobians of the B-spline’s value and its time derivatives with respect to an update of its parameters. For unit-length quaternion splines, the equations for angular velocity and angular acceleration are derived. An implementation of this algorithm is ...

Policy Learning with an Efficient Black-Box Optimization Algorithm

Robotic learning on real hardware requires an efficient algorithm which minimizes the number of trials needed to learn an optimal policy. Prolonged use of hardware causes wear and tear on the system and demands more attention from an operator. To this end, we present a novel black-box optimization algorithm, Reward Optimization with Compact Kernels and fast natural gradient regression (ROCK⋆). Our algorithm immediately updates knowledge after a single trial and is able to extrapolate in a controlled manner. These features make fast and safe learning on real hardware possible. The performance of our method is evaluated with standard benchmark functions that are commonly used to test optimization algorithms. We also present three different robotic optimization examples using ROCK⋆. The first robotic example is on a simulated robot arm, the second is on a real articulated legged system, and the third is on a simulated quadruped robot with 12 actuated joints. ROCK⋆ outperforms the current state-of-the-art algorithms in all tasks sometimes even by an order of magnitude.

Evaluation of Combined Time-Offset Estimation and Hand-Eye Calibration on Robotic Datasets

Using multiple sensors often requires the knowledge of static transformations between those sensors. If these transformations are unknown, hand-eye calibration is used to obtain them. Additionally, sensors are often unsynchronized, thus requiring time-alignment of measurements. This alignment can further be hindered by having sensors that fail at providing useful data over a certain time period. We present an end-to-end calibration framework to solve the hand-eye calibration. After an initial time-alignment step, we use the time-aligned pose estimates to perform the static transformation estimation based on different prefiltering methods, which are robust to outliers. In a final step, we employ a non-linear optimization to locally refine the calibration and time-alignment. Successful application of this estimation framework is demonstrated on multiple robotic systems with different sensor configurations. This framework is released as open source software together with the datasets.