ICRA 2018 Papers

Robotics controllers often consist of expert-designed heuristics, which can be hard to tune in higher dimensions. Simulation can aid in optimizing these controllers if parameters learned in simulation transfer to hardware. Unfortunately, this is often not the case in legged locomotion, necessitating learning directly on hardware. This motivates using data-efficient learning techniques like Bayesian Optimization (BO) to minimize collecting expensive data samples. BO is a black-box data-efficient optimization scheme, though its performance typically degrades in higher dimensions. We aim to overcome this problem by incorporating domain knowledge, with a focus on bipedal locomotion. In our previous work, we proposed a feature transformation that projected a 16-dimensional locomotion controller to a 1-dimensional space using knowledge of human walking. When optimizing a human-inspired neuromuscular controller in simulation, this feature transformation enhanced sample efficiency of BO over traditional BO with a Squared Exponential kernel. In this paper, we present a generalized feature transform applicable to non-humanoid robot morphologies and evaluate it on the ATRIAS bipedal robot, in both simulation and hardware. We present three different walking controllers and two are evaluated on the real robot. Our results show that this feature transform captures important aspects of walking and accelerates learning on hardware and simulation, as compared to traditional BO.

Developments for 3D control of the center of mass (CoM) of biped robots are currently located in two local minima: on the one hand, methods that allow CoM height variations but only work in the 2D sagittal plane; on the other hand, nonconvex direct transcriptions of centroidal dynamics that are delicate to handle. This paper presents an alternative that controls the CoM in 3D via an indirect transcription that is both low-dimensional and solvable fast enough for real-time control. The key to this development is the notion of boundedness condition, which quantifies the capturability of 3D CoM trajectories.

In Multi-Policy Decision-Making (MPDM), many computationally-expensive forward simulations are performed in order to predict the performance of a set of candidate policies. In risk-aware formulations of MPDM, only the worst outcomes affect the decision making process, and efficiently finding these influential outcomes becomes the core challenge. Recently, stochastic gradient optimization algorithms, using a heuristic function, were shown to be significantly superior to random sampling. In this paper, we show that accurate gradients can be computed - even through a complex forward simulation - using approaches similar to those in deep networks. We show that our proposed approach finds influential outcomes more reliably, and is faster than earlier methods, allowing us to evaluate more policies while simultaneously eliminating the need to design an easily-differentiable heuristic function. We demonstrate significant performance improvements in simulation as well as on a real robot platform navigating a highly dynamic environment.

UAVs must continuously adapt their mission to face unexpected internal or external hazards. This paper proposes a new BFM model (Bayesian Networks built from FMEA tables for MDP). This scalable model offers a modular and comprehensive method to incorporate different types of diagnosis modules based on BN (Bayesian Networks) and FMEA table (Failure Mode and Effects Analysis) to mission specifications expressed as a MDP (Markov Decision Processes). The BFM model implements the complete decision making process that covers both the application configurations at the embedded system level and the mission planning at the UAV level. These decisions are based on the QoS (Quality of Service) of applications, the resource use and the system and sensors health. We demonstrate on a case study for a target tracking mission that the BFM model can interface hazards and applications specifications and can improve the success and quality of the mission. To the best of our knowledge, this is the first proposal of a systematic method that integrates diagnosis modules to MDP model in order to take care of the implementation of embedded applications during a mission.

Soft robotics has advanced the field of biomedical engineering by creating safer technologies for interfacing with the human body. One of the challenges in this field is the realization of modular soft basic constituents and accessible assembly methods to increase the versatility of soft robots. We present a soft pneumatic actuator composed of two elastomeric strands that provide interdependent axial and radial expansion due to the modularity of the components and their helical arrangement. The actuator reaches 35% of elongation with respect to its initial height and both chambers achieve forces of 1N at about 19kPa. We describe the design, fabrication, modeling and benchtop testing of the soft actuator towards realizing 3D functional structures with potential medical applications. An example of application for soft medical robots is tissue regenerative for the long-gap esophageal atresia condition.

Fully autonomous navigation in populated environments is still a challenging problem for mobile robots. This paper explores the idea of using active human-robot communication to facilitate navigation tasks. We propose to convey a robot's intent to human users via a wearable haptic interface. The interface can display distinct haptic cues by modulating vibration amplitudes and patterns. We applied the concept to a single human/single robot orthogonal encounter scenario, where one of the two parties has to yield the right of way to avoid collision. Under certain conditions, the robot's intent (to yield to the human or not) is revealed to the human via the haptic interface prior to the interaction. We conducted an experiment with 10 users, in which the robot was teleoperated as a substitute for autonomy. Results show that, when given priority, users become more risk-accepting and use different strategies to navigate the collision scenario than when the robot takes priority or there is no haptic communication channel. In addition, we propose a social-force based model to predict human movement during navigation. The effect of communication can be explained as a shift in the user's safety buffer and expectation of the robot's future velocity.

For obstacles moving with high speeds, existing motion planning methods can rarely guarantee collision avoidance. This paper proposes a viable two-period velocity obstacle algorithm where one period predicts potential collisions within a limited time horizon, and the second period foresees collisions beyond that horizon. The second period is activated only when the obstacle's moving speed is larger than the maximum speed of the robot. The applicability of the new algorithm and the related computation issues are discussed. Both computer simulations and laboratory experiments illustrated the effectiveness of the proposed obstacle avoidance algorithm.

Cable-Driven Parallel Robots (CDPRs) have several advantages over conventional parallel manipulators most notably a large workspace. CDPRs whose workspace can be further increased by modification of the geometric architecture are known as Reconfigurable Cable Driven Parallel Robots(RCDPRs). A novel concept of RCDPRs, known as Mobile CDPR (MCDPR) that consists of a CDPR carried by multiple mobile bases, is studied in this paper. The system is capable of autonomously navigating to a desired location then deploying to a standard CDPR. In this paper, we analyze the Static equilibrium (SE) of the mobile bases when the system is fully deployed. In contrast to classical CDPRs we show that the workspace of the MCDPR depends, not only on the tension limits, but on the SE constraints as well. We demonstrate how to construct the Available Wrench Set (AWS) for a planar MCDPR wih a point-mass end-effector using both the convex hull and Hyperplane shifting methods. The obtained results are validated in simulation and on an experimental platform consisting of two mobile bases and a CDPR with four cables.

We present an autonomous and portable hip-only soft exosuit, for augmenting human walking and running that assists hip extension by delivering peak forces of 300N to the user. Different fixed assistance profiles for walking and running were applied based on an online classification algorithm. The approach is based on the biomechanical understanding that the center of mass potential energy fluctuations during walking and running are out of phase. Specifically, we monitor the vertical acceleration with an abdomen-mounted IMU at the moment of maximum hip extension. Validation is demonstrated with six subjects on the treadmill and with eight subjects outdoors. Our results demonstrated a 99.99% accuracy on average over the fourteen participants for various speeds (0.5 - 4m/s), slopes (-10 -20%), treadmill and overground terrain, loaded (13.6 kg) and unloaded, Exo On and Exo Off conditions, and different shoe types. Results from an evaluation outdoors overground on the energetics of eight subjects demonstrated a significant reduction for running when comparing Exo On to No Exo (3.9%) and for walking and running when comparing Exo On to Exo Off (12.2% and 8.2% respectively). This study represents the first demonstration of an autonomous wearable robot reducing the energy cost of running. Significant variation in response across subjects was observed, highlighting further improvements may be possible via assistance profile individualization with human-in-the-Ioop optimization.

State-of-the-art autonomous driving systems rely heavily on detailed and highly accurate prior maps. However, outside of small urban areas, it is very challenging to build, store, and transmit detailed maps since the spatial scales are so large. Furthermore, maintaining detailed maps of large rural areas can be impracticable due to the rapid rate at which these environments can change. This is a significant limitation for the widespread applicability of autonomous driving technology, which has the potential for an incredibly positive societal impact. In this paper, we address the problem of autonomous navigation in rural environments through a novel mapless driving framework that combines sparse topological maps for global navigation with a sensor-based perception system for local navigation. First, a local navigation goal within the sensor view of the vehicle is chosen as a waypoint leading towards the global goal. Next, the local perception system generates a feasible trajectory in the vehicle frame to reach the waypoint while abiding by the rules of the road for the segment being traversed. These trajectories are updated to remain in the local frame using the vehicle's odometry and the associated uncertainty based on the least-squares residual and a recursive filtering approach, which allows the vehicle to navigate road networks reliably, and at high speed, without detailed prior maps. We demonstrate the performance of the system on a full-scale autonomous vehicle navigating in a challenging rural environment and benchmark the system on a large amount of collected data.

Soft exosuits are a new approach for assisting with human locomotion, which applies assistive torques to the wearer through functional apparel. In this paper, we present a new version of autonomous multi-joint soft exosuit for gait assistance, particularly designed for overground walking. The soft exosuit assists with ankle plantarflexion, hip flexion, and hip extension, equally distributing the forces between ankle plantarflexion and hip flexion. A mobile actuation system was developed to generate high assistive forces, and Bowden cables are used to transmit the forces to the exosuit. A sensor harness connects two load cells and three IMU s per leg that are used to measure real-time data for a controller that commands desired force profiles as a function of the walking cycle. In addition, a control adaptation method was developed which adjusts control parameters while walking on irregular surfaces. In preliminary studies, the proposed method substantially improved the force consistency while walking over uneven terrain. Specifically, the number of steps where the peak force deviated from the target force decreased from 100 to 57 out of 250 steps, and RMS error on the peak force decreased from 90.0 N to 76.6 N with respect to 300 N target force. Also, a two-subject case study on country-course walking demonstrated the potential of this soft exosuit to improve human energy economy while walking overground.

Unmanned aerial vehicles (UAVs) are increasingly being proposed for a wide range of applications. A promising new class of these vehicles, known as agile fixed-wing UAV s, is intended to bridge the gap between conventional fixed-wing aircraft, which can cover long distances efficiently, and rotorcraft, which are typically very maneuverable. This paper addresses the implementation of a controller for agile UAVs, beginning with a hardware-in-the-loop (HIL) simulator, followed by testing on a real platform, both implemented on the Pixhawk microcontroller. We replace the Xplane physics engine used in the standard Pixhawk HIL with our own in-house Matlab/Simulink high-fidelity simulation of an agile UA V. The HIL simulator is found to provide substantial advantages in the transition from pure simulation to experimental testing. Once the controller is integrated into the flight platform, flight tests are conducted, and the results of those tests are compared to those from the HIL simulation and those obtained from the pure simulation environment, for maneuvers including hover, aggressive turnaround, knife-edge, and rolling Harrier. The desired position and orientation time histories were successfully tracked with the proposed implementation, demonstrating the impressive autonomous maneuverability that can be achieved by this type of aircraft.

Applications of robots for gathering data in underwater environments has been limited due to the challenges posed by the medium. We have developed a miniature, agile, easy to carry and deploy Autonomous Underwater Vehicle (AUV) equipped with a suite of sensors for underwater environmental sensing. We have also developed a compact high resolution fast temperature sensing module for the AUV for microstructure and turbulence measurements in water bodies. In this paper, we describe a number of algorithms and subsystems of the AUV that enable autonomous real-world operation, and present the data gathered in an experimental campaign in collaboration with limnologists. We demonstrate adaptive sampling missions where the AUV could autonomously locate a zone of interest and adapt its trajectory to stay in it. Further, it could execute specific behaviors to accommodate special sensing requirements necessary to enhance the quality of the data collected. In these missions, the AUV could autonomously trace a feature and capture horizontal variation in various quantities, including turbidity and temperature fluctuations, allowing limnologists to study lake phenomena in an additional dimension.

This paper presents a novel approach for autonomously mounting a sensor package on a vertical surface with an unmanned aerial vehicle (UAV). The Interacting-BoomCopter (I-BC) UAV uses an on-board webcam and computer along with a horizontally-mounted reversible propeller on its front boom to autonomously perform the aerial manipulation task. An overview of the vehicle design is presented along with the image processing algorithms used for target tracking, and the implementation of an extended finite state machine (EFSM) for carrying out the high-level autonomous control. The effectiveness of the autonomous control strategy and I-BC platform are examined through the performance of several autonomous sensor mounting flight tests.

Small-object detection and avoidance in unknown environments is a significant challenge to overcome for small autonomous vehicles that are generally highly agile and restricted in payload and computational processing power. Typical machine-vision and range measurement based solutions suffer either from restricted fields-of-view or significant computational complexity and are not easily portable to small platforms. In this paper, a novel bio-inspired navigation technique is introduced that is modeled using analogues of the small-field motion-sensitive interneurons of the insect visuomotor system. The proposed technique achieves small-field object detection based on Fourier residual analysis of instantaneous optic flow. The small field signal is used to extract relative range and bearing of the nearest obstacle, which is then combined with an artificial potential function-based low-order steering control law. The proposed sensing and control scheme is experimentally validated with a quadrotor vehicle that is able to effectively navigate an unknown environment laden with small-field clutter. This bio-inspired approach is computationally efficient and serves as a robust, reflexive solution to the problem of small-object detection and avoidance for autonomous robots.

This paper presents the autonomous battery exchange operation for small scale UAVs, using a mobile ground base that carries a robotic arm and a service station containing the battery exchange mechanism. The goal of this work is to demonstrate the means to increase the autonomy and persistence of robotic systems without requiring human intervention. The design and control of the system and its components are presented in detail, as well as the collaborative software framework used to plan and execute complex missions. Finally, the results of autonomous outdoor experiments are presented, in which the ground rover successfully localizes, retrieves, services, and deploys the landed UAV, proving its capacity to extend and enhance autonomous operations.

In this paper, an automatic planner for minimally invasive neurosurgery is presented. The solution provides the neurosurgeon with the best path to connect a user-defined entry point with a target in accordance with a specific cost function. The approach guarantees the avoidance of obstacles which can be found along the insertion pathway. The method is tailored to the EDEN2020* programmable bevel-tip needle, a multisegment steerable probe intended to be used to perform drug delivery for the treatment of glioblastomas. A sample-based heuristic search inspired by the BIT* algorithm is used to define the asymptotically-optimal solution in terms of path length, followed by a smoothing phase to meet the required kinematic constraints of the needle. To account for inaccuracies in catheter modeling, which could determine unexpected control errors over the insertion procedure, an uncertainty margin is defined in order to increase the algorithm's safety. The feasibility of the proposed solution was demonstrated by testing the method in simulated neurosurgical scenarios with different degrees of obstacle occupancy and against other sample-based algorithms present in literature: RRT, RRT* and an enhanced version of the RRT-Connect.

The estimation of the fabric material property during the folding is presented. The available techniques for the accurate garment folding rely on known material properties. Currently, the properties are estimated by an operator in advance of folding. We propose an iterative strategy, which updates the property while the garment is folded. The estimation is formulated as an optimisation task. It is based on measurements from a laser range finder. The proposed algorithm improves the estimation iteratively and prevents the garment from slipping at the same time. We demonstrate the estimation procedure for 10 fabric strips of different materials.

Vision systems are widely used in autonomous vehicle systems due to the rich information that camera sensors provide of the surrounding environment. This paper presents an automatic algorithm to obtain the drivable path of a vehicle operating in urban roads with or without clear lane markings. The developed system projects trajectories obtained during human operation of the vehicle and utilizes these to generate automatic labels for training a semantic based path prediction model. The system segments an urban scenario into 13 categories including vehicles, pedestrian, undrivable road, other categories relevant to urban roads, and a new class for a path proposal. The drivable path information is essential particularly in unstructured scenarios, and is critical for an intelligent vehicle system to make sound driving decisions. The path proposal category is a car-width drivable lane estimated to be safe to drive for the vehicle under consideration. The data collection, model training and inference process requires only images from a monocular camera and odometry from a low-cost IMU combined with a wheel encoder. The algorithm has been successfully demonstrated on the Sydney University campus, which is a challenging environment without clear road markings. The algorithm was demonstrated to run in real-time, proving its applicability for intelligent vehicles.