In previous work, model-based reinforcement learning was applied to a real-world labyrinth game to demonstrate sample-efficient learning using world models. In this paper, we further enhance sample efficiency and autonomy by introducing selective reconstruction: instead of reconstructing the full visual observation, our approach reconstructs only the low-dimensional physical state signals (e.g., marble position and plate inclination), while still leveraging the complete visual input for decision-making. This targeted reconstruction focuses the world model on learning dynamics-relevant information, thereby reducing computational overhead and model complexity. Additionally, we incorporate prioritized experience replay to accelerate learning in newly explored regions of the maze and implement an autonomous marble reloader to eliminate manual resets. Together, these enhancements reduce the required collected experience from 5 hours to 1.5 hours while achieving comparable performance, and enable fully autonomous learning without human supervision.

Distributed tactile sensing for multiforce detection is crucial for various aerial robot interaction tasks. However, current contact sensing solutions on drones only exploit single end-effector sensors and cannot provide distributed multicontact sensing. Designed to be easily mounted at the bottom of a drone, we propose an optical tactile sensor that features a large and curved soft-sensing surface, a hollow structure and a new illumination system. Even when spaced only 2 cm apart, multiple contacts can be detected simultaneously using our software pipeline, which provides real-world quantities of 3-D contact locations (mm) and 3-D force vectors (N), with an accuracy of 1.5 mm and 0.17 N, respectively. We demonstrate the sensor's applicability and reliability onboard and in real time with two demos related to, first, the estimation of the compliance of different perches and subsequent realignment and landing on the stiffer one, and second, the mapping of sparse obstacles. The implementation of our distributed tactile sensor represents a significant step toward attaining the full potential of drones as versatile robots capable of interacting with and navigating within complex environments.

Motivated by the challenge of achieving rapid learning in physical environments, this paper presents the development and training of a robotic system designed to navigate and solve a labyrinth game using model-based reinforcement learning techniques. The method involves extracting low-dimensional observations from camera images, along with a cropped and rectified image patch centered on the current position within the labyrinth, providing valuable information about the labyrinth layout. The learning of a control policy is performed purely on the physical system using model-based reinforcement learning, where the progress along the labyrinth’s path serves as a reward signal. Additionally, we exploit the system’s inherent symmetries to augment the training data. Consequently, our approach learns to successfully solve a popular real-world labyrinth game in record time, with only 5 hours of real-world training data.

Grasping objects whose physical properties are unknown is still a great challenge in robotics. Most solutions rely entirely on visual data to plan the best grasping strategy. However, to match human abilities and be able to reliably pick and hold unknown objects, the integration of an artificial sense of touch in robotic systems is pivotal. This paper describes a novel model-based slip detection pipeline that can predict possibly failing grasps in real-time and signal a necessary increase in grip force. As such, the slip detector does not rely on manually collected data, but exploits physics to generalize across different tasks. To evaluate the approach, a state-of-the-art vision-based tactile sensor that accurately estimates distributed forces was integrated into a grasping setup composed of a six degrees-of-freedom cobot and a two-finger gripper. Results show that the system can reliably predict slip while manipulating objects of different shapes, materials, and weights. The sensor can detect both translational and rotational slip in various scenarios, making it suitable to improve the stability of a grasp.

This paper presents an offset-free model predictive controller for fast and accurate control of a spherical soft robotic arm. In this control scheme, a linear model is combined with an online disturbance estimation technique to systematically compensate model deviations. Dynamic effects such as material relaxation resulting from the use of soft materials can be addressed to achieve offset-free tracking. The tracking error can be reduced by 35% when compared to a standard model predictive controller without a disturbance compensation scheme. The improved tracking performance enables the realization of a ball catching application, where the spherical soft robotic arm can catch a ball thrown by a human.

Data-driven approaches to tactile sensing aim to overcome the complexity of accurately modeling contact with soft materials. However, their widespread adoption is impaired by concerns about data efficiency and the capability to generalize when applied to various tasks. This paper focuses on both these aspects with regard to a vision-based tactile sensor, which aims to reconstruct the distribution of the three- dimensional contact forces applied on its soft surface. Accurate models for the soft materials and the camera projection, derived via state-of-the-art techniques in the respective domains, are employed to generate a dataset in simulation. A strategy is proposed to train a tailored deep neural network entirely from the simulation data. The resulting learning architecture is directly transferable across multiple tactile sensors without further training and yields accurate predictions on real data, while showing promising generalization capabilities to unseen contact conditions.

This paper presents a vision-based sensing approach for a soft linear actuator, which is equipped with an internal camera. The proposed vision-based sensing pipeline predicts the three-dimensional tip position of the actuator. To train and evaluate the algorithm, predictions are compared to ground truth data from an external motion capture system. An off-the-shelf distance sensor is integrated in a second actuator of the same type, providing only the vertical component of the tip position and used as a baseline for comparison. The camera-based sensing pipeline runs at 40 Hz in real-time on a standard laptop and is additionally used for closed loop elongation control of the actuator. It is shown that the approach can achieve comparable accuracy to the distance sensor for measuring the linear expansion of the actuator, but additionally provide the full three-dimensional tip position.

This paper presents the application of an iterative learning control scheme to improve the position tracking performance for an articulated soft robotic arm during aggressive maneuvers. Two antagonistically arranged, inflatable bellows actuate the robotic arm and provide high compliance while enabling fast actuation. Switching valves are used for pressure control of the soft actuators. A norm-optimal iterative learning control scheme based on a linear model of the system is presented and applied in parallel with a feedback controller. The learning scheme is experimentally evaluated on an aggressive trajectory involving set point shifts of 60 degrees within 0.2 seconds. The effectiveness of the learning approach is demonstrated by a reduction of the root-mean-square tracking error from 13 degrees to less than 2 degrees after applying the learning scheme for less than 30 iterations.

In recent years, ultra-wideband radio technology has become increasingly popular as a space-and cost-effective solution to the problem of indoor localization. This paper demonstrates how measurements of the channel impulse response can be used to estimate a signal's angle of arrival at a receiving antenna. This novel method requires no additional hardware, uses only a single antenna, and works with unsynchronized clocks and one-way communication. We evaluate our method on a real-dataset, and experimentally demonstrate how a mobile robot can localize itself by measuring angles to multiple ultra-wideband anchors.

Due to the complexity of modeling the elastic properties of materials, the use of machine learning algorithms is continuously increasing for tactile sensing applications. Recent advances in deep neural networks applied to computer vision make vision-based tactile sensors very appealing for their high-resolution and low cost. A soft optical tactile sensor that is scalable to large surfaces with arbitrary shape is discussed in this paper. A supervised learning algorithm trains a model that is able to reconstruct the normal force distribution on the sensor's surface, purely from the images recorded by an internal camera. In order to reduce the training times and the need for large datasets, a calibration procedure is proposed to transfer the acquired knowledge across multiple sensors while maintaining satisfactory performance.

A method for recovery from the complete loss of the state estimate is presented for autonomous quadcopters. Given an aerodynamic force model, the only measurements used to reinitialize the state estimate by means of a bank of extended Kalman filters are the angular rate and linear acceleration measurements of an IMU. The method is integrated within a complete recovery logic on a quadcopter platform and experimentally evaluated.

In this paper we present the design of a hybrid robotic arm using soft, inflatable bladders for actuation. Low cost switching valves are used for pressure control, where the valve model is identified experimentally. A model of the robotic arm is derived based on system identification and used to derive a linear quadratic Gaussian controller. A method to solve limitations of the employed switching valves is proposed and experimentally proven to improve tracking performance. The closed loop control performance of the robotic arm is demonstrated by stabilizing a rotational inverted pendulum known as the Furuta pendulum.

This paper presents a computationally efficient method of generating state-to-state trajectories for fully actuated multirotor vehicles. The approach consists of computing translational and rotational motion primitives that guide the vehicle from any initial state, defined by position, velocity, and attitude, to any end state in a given time, and subsequently verifying the motion primitives’ feasibility. Computationally lightweight motion primitives for which closed-form solutions exist are presented and an efficient method to test their feasibility is derived. The algorithm is shown to be able to generate trajectories and verify their feasibility within a few microseconds and can thus be used as an implicit feedback law or in high-level path planners that involve evaluating a large number of possible trajectories to achieve some high-level goal. The algorithm's performance is analyzed by comparing it with time-optimal trajectories. An experimental demonstration that requires the computation of trajectories for a large set of end states in real time is used to evaluate the approach.

This article discusses the implementation of an infinite-horizon model predictive control approach that is based on representing input and state trajectories by a linear combination of basis functions. An iterative constraint sampling strategy is presented for guaranteeing constraint satisfaction over all times. It will be shown that the proposed method converges. In addition, we will discuss the implementation of the resulting (online) model predictive control algorithm on an unmanned aerial vehicle and provide experimental results. The computational efficiency of the algorithm is highlighted by the fact that a sampling rate of 100 Hz was achieved on an embedded platform.

We present a global controller for tracking nominal trajectories with a flying wing tailsitter vehicle. The control strategy is based on a first-principles model of the vehicle dynamics that captures all relevant aerodynamic effects, and we apply an onboard parameter learning scheme in order to estimate unknown aerodynamic parameters. A cascaded control architecture is used: Based on position and velocity errors an outer control loop computes a desired attitude keeping the vehicle in coordinated flight, while an inner control loop tracks the desired attitude using a lookup table with precomputed optimal attitude trajectories. The proposed algorithms can be implemented on a typical microcontroller and the performance is demonstrated in various experiments.

In this paper we present the design and control of a novel six degrees-of-freedom aerial vehicle. Based on a static force and torque analysis for generic actuator configurations, we derive an eight-rotor configuration that maximizes the vehicle's agility in any direction. The proposed vehicle design possesses full force and torque authority in all three dimensions. A control strategy that allows for exploiting the vehicle's decoupled translational and rotational dynamics is introduced. A prototype of the proposed vehicle design is built using reversible motor-propeller actuators and capable of flying at any orientation. Preliminary experimental results demonstrate the feasibility of the novel design and the capabilities of the vehicle.

An approximate model predictive control approach is applied on an unmanned aerial vehicle with limited computational resources. A novel method using a continuous time parametrization of the state and input trajectory is used to derive a compact description of the optimal control problem. Different first order methods for the online optimization are discussed in terms of memory requirements and execution time. The generalized fast dual gradient method is implemented on the aerial vehicle. The approximate model predictive control algorithm runs on an embedded platform with a STM32 Cortex M4 processor. Simulation studies show that the model predictive controller outperforms a linear quadratic regulator in aggressive maneuvers. The model predictive control approach is evaluated in practice and shown to yield satisfactory flight behavior.

This paper presents the “monospinner”: a mechanically simple flying vehicle with only one moving part. The vehicle is shown to be controllable in three translational degrees of freedom and two rotational degrees of freedom. The vehicle has a single scalar control input, the thrust magnitude, and is controlled by a cascaded control strategy with an inner attitude controller and an outer position controller. The vehicle design is chosen based on two robustness metrics: the ability to maintain hover under perturbations and the probability of input saturation based on a stochastic model. The resulting mechanical and control designs are experimentally demonstrated, where it is also shown that the vehicle is sufficiently robust to achieve hover after being thrown into the air.

This paper proposes a generalized algorithm that enables mobile agents to meet in a bounded region based only on bearing information of other agents within their vicinity. Each agent repeatedly employs a stop-and-go strategy consisting of the following three actions: (1) Estimate the bearing of agents in its vicinity, (2) compute a target point based on the estimates, and (3) move to that target point. The motivation and case study example is a modular robot, the Distributed Flight Array, which we employ to validate the proposed algorithm.

A state estimator for a quadrocopter is presented, using measurements from an accelerometer, angular rate gyroscope, and a set of ultra-wideband ranging radios. The estimator uses an extended aerodynamic model for the quadrocopter, where the full 3D airspeed is observable through accelerometer measurements. The remaining quadrocopter states, including the yaw orientation, are rendered observable by fusing ultra-wideband range measurements, under the assumption of no wind. The estimator is implemented on a standard microcontroller using readily-available, low-cost sensors. Performance is experimentally investigated in a variety of scenarios, where the quadrocopter is flown under feedback control using the estimator output.

This paper addresses one of the fundamental tasks for the aerial assembly of tensile structures: aerial knot-tying. It presents a framework for representing and realizing knots with flying machines. A suitable representation of the knot topology is introduced taking into account the use of supporting elements and the characteristics of flying machines. This information is then translated into three-dimensional trajectories for the vehicle performing the aerial knot-tying task. Furthermore, preliminary results suggest that the quality of the resulting knot can be improved by the use of an iterative learning algorithm. Experiments are performed with quadrocopters to validate the proposed approach. An accompanying video shows the aerial knot-tying process.

A robot localization system is presented that enables a robot to estimate its position within some space by passively receiving ultra-wideband radio signals from fixed-position modules. Communication from the fixed-position modules is one-way, allowing the system to scale to multiple robots. Furthermore, the system's high position update rate makes it suitable to be used in a feedback control system, and enables the robot to track and perform high-speed, dynamic motions. This paper describes the algorithmic underpinnings of the system, discusses design decisions and their impact on the performance of the resulting localization, and highlights challenges faced during implementation. Performance of the localization system is experimentally verified through comparison with data from a motion-capture system. Finally, the system's application to robot self-localization is demonstrated through integration with a quadrocopter.

This paper presents a method that enables highspeed, steady flight in confined spaces for tethered quadrocopters. Thanks to the centripetal force exerted by the tether, high-speed trajectories along circles at different velocities, accelerations, and orientations in space can be flown. Various circular maneuvers are experimentally demonstrated, and tangential velocities of up to 15 m/s and centripetal accelerations of more than 13 g can be achieved in steady flight. The recorded data allows to characterize the flight behavior of quadrocopters at high airspeeds: As an example, an estimate of the actual thrust produced by the motors and of the aerodynamic drag acting on the vehicle is presented. An accompanying video shows tethered quadrocopters performing high-speed maneuvers.

In this paper, we present the design and implementation of Rapyuta, an open-source cloud robotics platform. Rapyuta helps robots to offload heavy computation by providing secured customizable computing environments in the cloud. The computing environments also allow the robots to easily access the RoboEarth knowledge repository. Furthermore, these computing environments are tightly interconnected, paving the way for deployment of robotic teams. We also describe three typical use cases, some benchmarking and performance results, and two proof-of-concept demonstrations. Note to Practitioners - Rapyuta allows to outsource some or all of a robot's onboard computational processes to a commercial data center. Its main difference to other, similar frameworks like the Google App Engine is that it is specifically tailored towards multiprocess high-bandwidth robotics applications/middlewares and provides a well-documented open-source implementation that can be modified to cover a large variety of robotic scenarios. Rapyuta supports the outsourcing of almost all of the current 3000+ ROS packages out of the box and is easily extensible to other robotic middleware. A pre-installed Amazon Machine Image (AMI) is provided that allows to launch Rapyuta in any of Amazon's data center within minutes. Once launched, robots can authenticate themselves to Rapyuta, create one or more secured computational environments in the cloud and launch the desired nodes/processes. The computing environments can also be arbitrarily connected to build parallel computing architectures on the fly. The WebSocket-based communication protocol, which provides synchronous and asynchronous communication mechanisms, allows not only ROS based robots, but also browsers and mobiles phones to connect to the ecosystem. Rapyuta's computing environments are private, secure, and optimized for data throughput. However, its performance is in large part determined by the latency and quality of the network connection and the performance of the data center. Optimizing performance under these constraints is typically highly application-specific. The paper illustrates an example of performance optimization in a collaborative real-time 3-D mapping application. Other target applications include collaborative 3-D mapping, task/grasp planning, object recognition, localization, and teleoperation, among others.

This paper presents an architecture, protocol, and parallel algorithms for collaborative 3D mapping in the cloud with low-cost robots. The robots run a dense visual odometry algorithm on a smartphone-class processor. Key-frames from the visual odometry are sent to the cloud for parallel optimization and merging with maps produced by other robots. After optimization the cloud pushes the updated poses of the local key-frames back to the robots. All processes are managed by Rapyuta, a cloud robotics framework that runs in a commercial data center. This paper includes qualitative visualization of collaboratively built maps, as well as quantitative evaluation of localization accuracy, bandwidth usage, processing speeds, and map storage.

This paper presents a trajectory generation algorithm that efficiently computes high-performance flight trajectories that are capable of moving a quadrocopter from a large class of initial states to a given target point that will be reached at rest. The approach consists of planning separate trajectories in each of the three translational degrees of freedom, and ensuring feasibility by deriving decoupled constraints for each degree of freedom through approximations that preserve feasibility. The presented algorithm can compute a feasible trajectory within tens of microseconds on a laptop computer; remaining computation time can be used to iteratively improve the trajectory. By replanning the trajectory at a high rate, the trajectory generator can be used as an implicit feedback law similar to model predictive control. The solutions generated by the algorithm are analyzed by comparing them with time-optimal motions, and experimental results validate the approach.

A method is presented for the rapid generation and feasibility verification of motion primitives for quadrocopters and similar multirotor vehicles. The motion primitives are defined by the quadrocopter's initial state, the desired motion duration, and any combination of components of the quadrocopter's position, velocity, and acceleration at the motion's end. Closed-form solutions for the primitives are given, which minimize a cost function related to input aggressiveness. Computationally efficient tests are presented to allow for rapid feasibility verification. Conditions are given under which the existence of feasible primitives can be guaranteed a priori . The algorithm may be incorporated in a high-level trajectory generator, which can then rapidly search over a large number of motion primitives which would achieve some given high-level goal. It is shown that a million motion primitives may be evaluated and compared per second on a standard laptop computer. The motion primitive generation algorithm is experimentally demonstrated by tasking a quadrocopter with an attached net to catch a thrown ball, evaluating thousands of different possible motions to catch the ball.

This paper presents periodic solutions for a quadrocopter maintaining a height around a position in space despite having lost a single, two opposing, or three propellers. In each case the control strategy consists of the quadrocopter spinning about a primary axis, fixed with respect to the vehicle, and tilting this axis for translational control. A linear, timeinvariant description of deviations from the attitude equilibrium is derived, allowing for a convenient cascaded control design. The results for the cases of losing one and two propellers are validated in experiment, while the case of losing three propellers is validated in a nonlinear simulation. These results have application in multicopter fault-tolerant control design, and also point to possible design directions for novel flying vehicles.

We present an iterative learning scheme for improving the performance of highly dynamic open-loop maneuvers with quadrocopters. A probabilistic estimate of the state deviation at the end of the maneuver is obtained by fusing two data sources that are available on-board: 1) an inertial measurement unit, and 2) control inputs from an external pilot that performs a recovery after the open-loop maneuver has been executed. A computationally lightweight policy gradient method is applied in order to adapt a set of characteristic maneuver parameters, which in turn reduces the expected value of the final state deviation for the next execution of the maneuver. The performance of the learning algorithm is demonstrated in the ETH Zurich Flying Machine Arena by improving the performance of a triple flip.

Presents an editorial opinion of drone technology. Drones, autonomous or teleoperated flying machines, have been an active area of research for decades. This guest editorial discusses the technology and applications that currently are supported by drones and assesses the future for the use and deploying of drones in the marketplace. Analyzes the benefits of deploying drones, its power system capacity, the economics of drone technology, and applications for future use.

In this paper we present the design and implementation of Rapyuta1, the RoboEarth Cloud Engine. Rapyuta is an open source Platform-as-a-Service (PaaS) framework designed specifically for robotics applications. Rapyuta helps robots to offload heavy computation by providing secured customizable computing environments in the cloud. The computing environments also allow robots to easily access the RoboEarth knowledge repository. Furthermore, these computing environments are tightly interconnected, paving the way for deployment of robotic teams. We also describe specific use case configurations and present some performance results.

This paper presents the building of lightweight tensile structures with quadrocopters. The construction elements (such as ropes, cables, and wires) in this kind of structure are subject to tension forces. This paper identifies the basic building elements (nodes, links) required for the construction of tensile structures, and translates them into meaningful trajectories for quadrocopters. The use of a library of building elements is suggested. Hybrid force-position control strategies based on admittance control are exploited. Prototypical tensile structures are built by quadrocopters to validate the proposed approach. An accompanying video shows the building process.

We present the design of a system that allows quadrocopters to balance an inverted pendulum, throw it into the air, and catch and balance it again on a second vehicle. Based on first principles models, a launch condition for the pole is derived and used to design an optimal trajectory to throw the pole towards a second quadrocopter. An optimal catching instant is derived and the corresponding position is predicted by simulating the current position and velocity estimates forward in time. An algorithm is introduced that generates a trajectory for moving the catching vehicle to the predicted catching point in real time. By evaluating the pole state after the impact, an adaptation strategy adapts the catch maneuver such that the pole rotates into the upright equilibrium by itself. Experimental results demonstrate the performance of the system.

Iterative Learning Control algorithms are based on the premise that “practice makes perfect”. By iteratively performing an action, repetitive errors can be learned and accounted for in subsequent iterations, in a non-causal and feedforward manner. This method has been previously implemented for a quadrocopter system, enabling the quadrocopter to learn to accurately track high-performance slalom trajectories. However, one major limitation of this system is that knowledge from previously learned trajectories is not generalized or transferred to new trajectories; these must be learned from a state of zero experience. This paper experimentally shows that the major dynamics of the Iterative Learning Control process can be captured by a linear map, trained on previously learned slalom trajectories. This map enables this prior knowledge to be used to improve the initialization of an unseen trajectory. Experimental results show that prediction based on a single prior is enough to reduce the initial tracking error for an unseen trajectory by an order of magnitude.

Quadrocopters exhibit complex high-speed fight dynamics, and the accurate modeling of these dynamics has proven diffcult. Due to the use of simplifed models in the design of feedback control algorithms, the execution of highperformance fight maneuvers under pure feedback control typically leads to large tracking errors. This paper investigates an iterative learning scheme aimed at the non-causal compensation of repeatable trajectory tracking errors over the course of multiple executions of periodic maneuvers. The learning is carried out in the frequency domain and uses a simplifed model of the closed-loop dynamics of quadrocopter and feedback controller. The resulting algorithm requires little computational power and memory, and its convergence is shown for the nominal model. This paper further introduces a time-scaling method that allows the initial learning to occur at reduced speeds, thus extending the applicability of the algorithm for high performance maneuvers. The presented algorithms are validated in experiments, with a quadrocopter fying a fgure-eight maneuver at high speed.

This paper discusses distributed average consensus in the context of a distributed embedded system with multiple agents connected through a communication network. Adversities such as switching of network topologies, agents joining or leaving the network, and communication link creation or failure may arise in these systems. To address these difficulties, we propose an asynchronous implementation of a distributed average consensus algorithm that has the following properties: (1) unbiased average, (2) homogeneous implementation, (3) robustness to network adversities, (4) dynamic consensus, and (5) well-defined tuning parameters. We demonstrate an application of the implementation on a specific distributed embedded system, the Distributed Flight Array, where we solve two average consensus problems to estimate altitude and tilt of the vehicle from multiple distance measurements.

Given a hover-capable flying vehicle attached to a fixed point by a taut tether, we present a novel method to recover the vehicle's relative position and absolute orientation. The proposed method requires only on-board inertial sensors, and indirectly measures the string force, enabling the additional use of the tether as a physical user interaction medium. We present the vertical-plane dynamics of such a system and the localization approach, discuss sensitivity issues, and implement an estimator and controller based on the presented model. We demonstrate the method experimentally on a tethered quadrocopter in the Flying Machine Arena, using both a vertical-plane-constrained vehicle and in 3D.

A robot testbed for writing Chinese and Japanese calligraphy characters is presented. Single strokes of the calligraphy characters are represented in a database and initialized with a scanned reference image and a manually chosen initial drawing spline. A learning procedure uses visual feedback to analyze each new iteration of the drawn stroke and updates the drawing spline such that every subsequent drawn stroke becomes more similar to the reference image. The learning procedure can be performed either in simulation, using a simple brush model to create simulated images of the strokes, or with a real robot arm equipped with a calligraphy brush and a camera that captures images of the drawn strokes. Results from both simulations and experiments with the robot arm are presented.

An algorithm is proposed allowing for the rapid generation and evaluation of quadrocopter state interception trajectories. These trajectories are from arbitrary initial states to final states defined by the vehicle position, velocity and acceleration with a specified end of time. Sufficient criteria are then derived allowing trajectories to be tested for feasibility with respect to thrust and body rates. It is also shown that the range of a linear combination of the vehicle state can be solved for in closed form, useful e.g. for testing that the position remains within a box. The algorithm is applied by revisiting the problem of finding a trajectory to hit a ball towards a target with a racket attached to a quadrocopter. The trajectory generator is used in a model predictive control like strategy, where thousands of trajectories are generated and evaluated at every controller update step, with the first input of the optimal trajectory being sent to the vehicle. It is shown that the method can generate and evaluate on the order of one million trajectories per second on a standard laptop computer.

This paper introduces a method for carrying a flexible payload with multiple attached flying vehicles. A model for a particular class of flexible structures is presented, and an estimator is derived that observes the pose of the structure in space as well as the magnitude of some characteristic deformation modes. A control strategy that controls the flexible payload to a desired pose while also controlling the deformations to zero is introduced. The presented methods are validated in the ETH Zurich Flying Machine Arena by flying with a thin, flexible ring that is carried by six quadrocopters.

The key components of feedback control systems -- sensors, actuators, computation, power, and communication -- are continually becoming smaller, lighter, more robust, higher performance, and less expensive. By using appropriate algorithms and system architectures, it is thus becoming possible to "close the loop" on almost any machine, and to create new capabilities that fully exploit their dynamic potential. In this talk I will discuss various projects -- involving mobile robots, flying machines, an autonomous table, and actuated wingsuits -- where these new machine competencies are interfaced with the ultimate dynamic entities: human beings.

This paper presents an algorithm that generates collision-free trajectories in three dimensions for multiple vehicles within seconds. The problem is cast as a non-convex optimization problem, which is iteratively solved using sequential convex programming that approximates non-convex constraints by using convex ones. The method generates trajectories that account for simple dynamics constraints and is thus independent of the vehicle's type. An extensive a posteriori vehicle-specific feasibility check is included in the algorithm. The algorithm is applied to a quadrocopter fleet. Experimental results are shown.

This paper introduces the Cubli, a 15×15×15 cm cube that can jump up and balance on a corner. Momentum wheels mounted on three faces of the cube (Fig. 1) rotate at high angular velocities and then brake suddenly, causing the Cubli to jump up. Once the Cubli has almost reached the corner stand-up position, controlled motor torques are applied to make it balance on its corner. This paper tracks the development of the Cubli's one dimensional prototype at ETH Zurich and presents preliminary results.

This paper presents an algorithm that permits the calculation of interception maneuvers for quadrocopters. The translational degrees of freedom of the quadrocopter are decoupled. Pontryagin's minimum principle is used to show that the interception maneuver that minimizes the time to rest after the interception is identical to the time-optimal maneuver that drives the vehicle to the position at which it comes to rest after the interception. This fact is leveraged to apply previously developed, computationally efficient methods for the computation of interception maneuvers. The resulting trajectory generation algorithm is computationally lightweight, permitting its use as an implicit feedback law by replanning the trajectory at each controller update. The validity and performance of the approach is demonstrated experimentally by intercepting balls mid-flight. The real-time trajectory generation permits to take into account changes in the predicted ball flight path at each controller update.

Although thousands of Chinese characters exist, they can be constructed from a limited number of single strokes. In Chinese calligraphy these strokes are combined into a full character in a fluid way. Therefore Chinese calligraphy provides an interesting problem to study learning mechanisms such as how to automatically construct complex tasks (full characters) from previously learned simpler ones (single strokes) (Fig. 1). The goal of this project is that a robot should be able to decide which previously learned strokes or characters to use for drawing a newly presented character and to improve its drawing over several iterations.

This paper describes a generalized method for computing the altitude and attitude of a rigid body with respect to an inertial frame using a set of distance measurements obtained from a sensor network. In the case where all sensors are centrally measurable, a linear-optimal estimate is obtained. This method is used as a way for estimating altitude and attitude of the Distributed Flight Array, a modular multi-propeller flying vehicle where each module in the array obtains its own distance measurement and coordinates with its immediate neighbour(s) actions for flight. To account for communication bandwidth constraints, a scalable, distributed scheme is presented where each module shares local information. In the limit of sharing information, each module asymptotically computes the linear-optimal altitude and attitude estimate.

An autonomous safety mechanism is presented, as implemented in an indoor flying vehicle research testbed. The safety mechanism relies on integration of onboard gyroscope measurements and thrust commands to estimate the vehicle state for short lengths of time. It is used in the case of loss of external control signal or loss of external measurement data, to reduce the likelihood of a vehicle crash, or at least reduce the severity of an unavoidable crash. As UAVs move into ever more mainstream applications with increased public interaction, such safety systems become more critical.

We revisit a recently developed iterative learning algorithm that enables systems to learn from a repeated operation with the goal of achieving high tracking performance of a given trajectory. The learning scheme is based on a coarse dynamics model of the system and uses past measurements to iteratively adapt the feed-forward input signal to the system. The novelty of this work is an identification routine that uses a numerical simulation of the system dynamics to extract the required model information. This allows the learning algorithm to be applied to any dynamic system for which a dynamics simulation is available (including systems with underlying feedback loops). The proposed learning algorithm is applied to a quadrocopter system that is guided by a trajectory-following controller. With the identification routine, we are able to extend our previous learning results to three-dimensional quadrocopter motions and achieve significantly higher tracking accuracy due to the underlying feedback control, which accounts for non-repetitive noise.

Designing a controller that is scalable, robust, and that can adapt to an arbitrary configuration is one of the major challenges of modular robotics. This paper takes one step forward in addressing this challenge by presenting a methodology for controlling any flight-feasible configuration of a modular flying vehicle, in this case the Distributed Flight Array (DFA). In this work we present a well-structured, parameterized controller and describe a method for optimizing its parameters in order to achieve the best possible performance subject to the system's physical constraints. We then show how the configuration space of the DFA can be parameterized by only a few variables and propose a straightforward approach for mapping this configuration space to its control parameter space.

This paper presents a method for enabling a fleet of circularly arranged quadrocopters to throw and catch balls with a net. Based on a first-principles model of the net forces, nominal inputs for all involved vehicles are derived for arbitrary target trajectories of the net. Two algorithms that generate open-loop trajectories for throwing and catching a ball are also introduced. A set of throws and catches is demonstrated in the ETH Zurich Flying Machine Arena testbed.

The author reflects on the automation and robotics innovations that Kiva Systems developed that allow the company to successfully develop and deploy warehouse systems with hundreds - sometimes thousands - of autonomous mobile robots, and to describe some of the remaining research questions.

We present the design of the Blind Juggler: a robot that is able to juggle an unconstrained ball without feedback at heights of up to 2 m. The robot actuates a parabolic aluminum paddle with a linear motor. We achieve open-loop stability of the ball trajectory with two design parameters: 1) the curvature of the parabolic paddle and 2) the acceleration of the paddle at impact. We derive a linear map of perturbations of the nominal ball trajectory over a single bounce and obtain local stability of the trajectory by tuning the eigenvalues of this mapping with the two design parameters. We consider nine ball states in this analysis, including ball spin. Experimental data provide the impact states of the ball and paddle. From these data, we can identify system parameters and infer the process noise introduced into the system. We then combine the experimental noise power spectral densities with a model of the system and optimize the design parameters such that the impact of the process noise on juggling performance is minimized. Theoretical as well as experimental results of the optimization are discussed.

We extend the classic control problem of the inverted pendulum by placing the pendulum on top of a quadrotor aerial vehicle. Both static and dynamic equilibria of the system are investigated to find nominal states of the system at standstill and on circular trajectories. Control laws are designed around these nominal trajectories. A yaw-independent description of quadrotor dynamics is introduced, using a `Virtual Body Frame'. This allows for the time-invariant description of curved trajectories. The balancing performance of the controller is demonstrated in the ETH Zurich Flying Machine Arena testbed. Development potential for the future is highlighted, with a focus on applying learning methodology to increase performance by eliminating systematic errors that were seen in experiments.

The Flying Machine Arena (FMA) is an indoor research space built specifically for the study of autonomous systems and aerial robotics. In this video, we give an overview of this testbed and some of its capabilities. We show the FMA infrastructure and hardware, which includes a fleet of quadrocopters and a motion capture system for vehicle localization. The physical components of the FMA are complemented by specialized software tools and components that facilitate the use of the space and provide a unified framework for communication and control. The flexibility and modularity of the experimental platform is highlighted by various research projects and demonstrations.

We present the analysis and design process of the Pendulum Juggler, a robot that juggles a ball with an actuated paddle mounted to a swinging pendulum. The unconstrained ball travels up to 1 m horizontally, reaching an apex height of 1.1 m between impacts. Using a perturbation analysis, we show that the ball trajectory is locally stable without feedback. The pendulum consists of a four-bar linkage, optimized to keep the paddle perpendicular to the impact velocity of the ball. We verify the stability in experiments with a prototype of the Pendulum Juggler, demonstrating sustained juggling at amplitudes of up to 25°.

This paper presents a method allowing a quadrocopter with a rigidly attached racket to hit a ball towards a target. An algorithm is developed to generate an open loop trajectory guiding the vehicle to a predicted impact point - the prediction is done by integrating forward the current position and velocity estimates from a Kalman filter. By examining the ball and vehicle trajectories before and after impact, the system estimates the ball's drag coefficient, the racket's coefficient of restitution and an aiming bias. These estimates are then fed back into the system's aiming algorithm to improve future performance. The algorithms are implemented for three different experiments: a single quadrocopter returning balls thrown by a human; two quadrocopters co-operatively juggling a ball back-and-forth; and a single quadrocopter attempting to juggle a ball on its own. Performance is demonstrated in the Flying Machine Arena at the ETH Zurich.

A numerical method for computing quadrocopter maneuvers between two states is presented. Computed maneuvers satisfy Pontryagin's minimum principle with respect to time-optimality. First, in order to obtain the structure of time-optimal maneuvers, we apply the minimum principle to a first-principles, two-dimensional quadrotor model. Then we present a numerical algorithm that enables the computation of maneuvers for arbitrary initial and final states. The developed method is used to compute a set of maneuvers, which are discussed and demonstrated experimentally in the ETH Zurich Flying Machine Arena testbed.

We describe a simple and intuitive policy gradient method for improving parametrized quadrocopter multi-flips by combining iterative experiments with information from a first-principles model. We start by formulating an N-flip maneuver as a five-step primitive with five adjustable parameters. Optimization using a low-order first-principles 2D vertical plane model of the quadrocopter yields an initial set of parameters and a corrective matrix. The maneuver is then repeatedly performed with the vehicle. At each iteration the state error at the end of the primitive is used to update the maneuver parameters via a gradient adjustment. The method is demonstrated at the ETH Zurich Flying Machine Arena testbed on quadrotor helicopters performing and improving on flips, double flips and triple flips.

This paper introduces the Distributed Flight Array which is being developed at ETH Zurich. This multi-propeller platform consists of autonomous single-propeller modules that are able to drive, dock with their peers, and fly in a coordinated fashion. These modules are organized as distributed computational units with minimal sensory input. This is a complex system that is rich in dynamics with plenty of room to explore various distributed estimation and control strategies. Experimental demonstrations in docking, driving, and flight have proven its feasibility.

This paper presents a quadrocopter flying in rhythm to music. The quadrocopter performs a periodic side-to-side motion in time to a musical beat. Underlying controllers are designed that stabilize the vehicle and produce a swinging motion. Synchronization is then achieved by using concepts from phase-locked loops. A phase comparator combined with a correction algorithm eliminate the phase error between the music reference and the actual quadrocopter motion. Experimental results show fast and effective synchronization that is robust to sudden changes in the reference amplitude and frequency. Changes in frequency and amplitude are tracked precisely when adding an additional feedforward component, based on an experimentally determined look-up table.

An estimation algorithm is developed for determining pitch and roll angles (tilt) of a rigid body fixed at a pivot point using multiple accelerometers. The estimate is independent of the rigid body dynamics; the method is applicable both in static conditions and for any dynamic motion of the body. No dynamic model is required for the estimator; only the mounting positions of the sensors need to be known. The proposed estimator is the optimal linear estimate in a least-squares sense if knowledge of the system dynamics is not used. The estimate may be used as a basis for further filtering and fusion techniques, such as sensor fusion with rate gyro data. The estimation algorithm is applied to the problem of state estimation for the Balancing Cube, a rigid structure that can actively balance on its corners. Experimental results are provided.

We analyzed age-related changes in motor response in a visuomotor compensatory tracking task. Subjects used a manipulandum to attempt to keep a displayed cursor at the center of a screen despite random perturbations to its location. Cross-correlation analysis of the perturbation and the subject response showed no age-related increase in latency until the onset of response to the perturbation, but substantial slowing of the response itself. Results are consistent with age-related deterioration in the ratio of signal to noise in visuomotor response. The task is such that it is tractable to use Bayesian and quadratic optimality assumptions to construct a model for behavior. This model assumes that behavior resembles an optimal controller subject to noise, and parametrizes response in terms of latency, willingness to expend effort, noise intensity, and noise bandwidth. The model is consistent with the data for all young (n = 12, age 20-30) and most elderly (n = 12, age 65-92) subjects. The model reproduces the latency result from the cross-correlation method. When presented with increased noise, the computational model reproduces the experimentally observed age-related slowing and the observed lack of increased latency. The model provides a precise way to quantitatively formulate the long-standing hypothesis that age-related slowing is an adaptation to increased noise.

This paper presents an algorithm to iteratively drive a system quickly from one state to another. A simple model which captures the essential features of the system is used to compute the reference trajectory as the solution of an optimal control problem. Based on a lifted domain description of that same model an iterative learning controller is synthesized by solving a linear least-squares problem. The non-causality of the approach makes it possible to anticipate recurring disturbances. Computational requirements are modest, allowing controller update in real-time. The experience gained from successful maneuvers can be used to significantly reduce transients when performing similar motions. The algorithm is successfully applied to a real quadrotor unmanned aerial vehicle. The results are presented and discussed.

We describe the design of a juggling robot that is able to vertically bounce a completely unconstrained ball without any sensing. The robot consists of a linear motor actuating a machined aluminum paddle. The curvature of this paddle keeps the ball from falling off while the apex height of the ball is stabilized by decelerating the paddle at impact. We analyze the mapping of perturbations of the nominal trajectory over a single bounce to determine the design parameters that stabilize the system. The first robot prototype confirms the results from the stability analysis and exhibits substantial robustness to perturbations in the horizontal degree of freedoms. We then measure the performance of the robot and characterize the noise introduced into the system as white noise. This allows us to refine the design parameters by minimizing the H2norm of an input-output representation of the system. Finally, we design an H2optimal controller for the apex height using impact time measurements as feedback and show that the closed-loop performance is only marginally better than what is achieved with open-loop control.

We introduce patch models, a computational modeling formalism for multivehicle combat domains, based on spatiotemporal abstraction methods developed in the computer science community. The framework yields models that are expressive enough to accommodate nontrivial controlled vehicle dynamics while being within the representational capabilities of common artificial intelligence techniques used in the construction of autonomous systems. The framework allows several key design requirements of next-generation network-centric command and control systems, such as maintenance of shared situation awareness, to be achieved. Major features include support for multiple situation models at each decision node and rapid mission plan adaptation. We describe the formal specification of patch models and our prototype implementation, i.e., Patchworks. The capabilities of patch models are validated through a combat mission simulation in Patchworks, which involves two defending teams protecting a camp from an enemy attacking team.

Mixed-integer linear programming (MILP) is a powerful tool for planning and control problems because of its modeling capability and the availability of good solvers. However, for large models, MILP methods suffer computationally. In this paper, we present iterative MILP algorithms that address this issue. We consider trajectory-generation problems with obstacle-avoidance requirements and minimum-time trajectory-generation problems. These problems involve vehicles that are described by mixed logical dynamical equations, a form of hybrid system. The algorithms use fewer binary variables than standard MILP methods, and require less computational effort.

Describes the issues in robotics from a viewpoint of edutainment through a series of activities in the Robot World Cup Initiative and related events, such as the International Robot Games Festival (Robofesta) supported by the Japanese government to promote creative and imaginative education programs, RoboCup Jr. which is designed for kids and the younger generation to play RoboCup games with easily constructible platforms, development of small legged robots for pets in the house or games, and education projects in system engineering. Finally, concluding remarks for future activities are given.