IROS 2019 Papers

Mixing is important for microfluidic systems and placing obstacles in a flow path for generating advection, or splitting/recombination flows, is a very popular method for enhancing the mixing. In this paper, we present experimental investigations on such microfluidic mixing with four different shapes of trapezoidal obstacles and tested them in a 1000-fold span of Reynolds number (Re). The obstacles have four different base ratios 1, 3, 6 and 9, which indicate the ratios between the upper and lower bases of a trapezoid. Two different dyes, yellow and blue, were used for visualizing and evaluating the performance of mixing. The results show that, when the base ratio is 1, the mixing surprisingly becomes worse with the increase of the flowrate. When the ratio is 3 or 6, the mixing performance is first reduced with the increase of Re and then went through a transition around Re of 10 before it rapidly increases. When the ratio is 9, two transition points were found around Re of 10 and 100, respectively. The work provides useful information for realizing shape-dependent mixing performance, as well as for developing an intelligent lab-on-a-chip system.

A plethora of state estimation techniques have appeared in the last decade using visual data, and more recently with added inertial data. Datasets typically used for evaluation include indoor and urban environments, where supporting videos have shown impressive performance. However, such techniques have not been fully evaluated in challenging conditions, such as the marine domain. In this paper, we compare ten recent open-source packages to provide insights on their performance and guidelines on addressing current challenges. Specifically, we selected direct and indirect methods that fuse camera and Inertial Measurement Unit (IMU) data together. Experiments are conducted by testing all packages on datasets collected over the years with underwater robots in our laboratory. All the datasets are made available online.

Acquiring new robot motor skills is cumbersome, as learning a skill from scratch and without prior knowledge requires the exploration of a large space of motor configurations. Accordingly, for learning a new task, time could be saved by restricting the parameter search space by initializing it with the solution of a similar task. We present a framework which is able of such knowledge transfer from already learned movement skills to a new learning task. The framework combines probabilistic movement primitives with descriptions of their effects for skill representation. New skills are first initialized with parameters inferred from related movement primitives and thereafter adapted to the new task through relative entropy policy search. We compare two different transfer approaches to initialize the search space distribution with data of known skills with a similar effect. We show the different benefits of the two knowledge transfer approaches on an object pushing task for a simulated 3-DOF robot. We can show that the quality of the learned skills improves and the required iterations to learn a new task can be reduced by more than 60% when past experiences are utilized.

Plan execution on real robots in realistic environments is underdetermined and often leads to failures. The choice of action parameterization is crucial for task success. In this paper, we present a mechanism for a robot that is acting in a real-world environment to think ahead of time with fast plan projection and, thereby, choose action parameterizations that are predicted to lead to successful execution. For finding causal relationships between action parameterizations and task success, we provide the robot with means for plan introspection and propose a systematic and hierarchical plan structure to support that. We evaluate our approach by showing how a PR2 robot, when equipped with the proposed system, is able to choose action parameterizations that increase task execution success rates and overall performance of fetch and place actions in a real world setting.

We developed a system to evaluate the skill of operating a hydraulic excavator using a remotely controlled (RC) excavator and virtual reality (VR) technology. We re-modeled the RC excavator such that it can be operated in the same manner as an actual excavator, and proceeded to measure the excavator's state. To evaluate the skill of operating this system, we calculated several indices from data recorded during excavation work. We calculated the same indices from data recorded during excavation performed by an actual excavator, and verified that there exists a high correlation between the indices of the RC excavator and those of the actual excavator.

The employment of Social Assistive Robots (SARs) for monitoring elderly users represents a valuable gateway for at-home assistance. Their deployment in the house of the users can provide effective opportunities for early detection of Mild Cognitive Impairment (MCI), a condition of increasing impact in our aging society, by means of digitalized cognitive tests. In this work, we present a system where a specific set of cognitive tests is selected, digitalized, and integrated with a robotic assistant, whose task is the guidance and supervision of the users during the completion of such tests. The system is then evaluated by means of an experimental study involving potential future users, in order to assess its acceptability and identify key directions for technical improvements.

We present an end-to-end deep learning approach for performing metric scale-sensitive regression tasks such visual odometry with a single camera and no additional sensors. We propose a novel 3D convolutional architecture, 3DC-VO, that can leverage temporal relationships over a short moving window of images to estimate linear and angular velocities. The network makes local predictions on stacks of images that can be integrated to form a full trajectory. We apply 3DC-VO to the KITTI visual odometry benchmark and the task of estimating a pilot’s control inputs from a first-person video of a quadrotor flight. Our method exhibits increased accuracy relative to comparable learning-based algorithms trained on monocular images. We also show promising results for quadrotor control input prediction when trained on a new dataset collected with a UAV simulator.

Search-based planning for relatively low-dimensional motion-planning problems such as for autonomous navigation and autonomous flight has been shown to be very successful. Such framework relies on laying a grid over a state-space and constructing a set of actions (motion primitives) that connect the centers of cells. However, in some cases such as kinodynamic motion planning, planning for bipedal robots with high balance requirements, computing these actions can be highly non-trivial and often impossible depending on the dynamic constraints. In this paper, we explore a soft version of discretization, wherein the state-space remains to be continuous but the search tries to avoid exploring states that are likely to be duplicates of states that have already been explored. We refer to this property of the search as soft duplicate detection and view it as a relaxation of the standard notion of duplicate detection. Empirically, we show that the search can efficiently compute paths in highly-constrained settings and outperforms alternatives on several domains.

Designing efficient control strategies is well studied. Due to recent technological advancements and applications to the field of robotics, exploring ways to design optimal control for multi robot systems is gaining interest. In this respect, ergodicity has been successfully applied as an effective control technique for tracking and coverage. Generating flocking behaviour is a problem with many interesting special cases that include both tracking and coverage. The main contribution of this paper is the application of ergodicity to emulate flocking behaviour. Our approach is appealing, as control and communication are all assumed to be local, and so this behaviour is self-organized. Simulation results show that the proposed approach is effective.

Many modern nonlinear control methods aim to endow systems with guaranteed properties, such as stability or safety, and have been successfully applied to the domain of robotics. However, model uncertainty remains a persistent challenge, weakening theoretical guarantees and causing implementation failures on physical systems. This paper develops a machine learning framework centered around Control Lyapunov Functions (CLFs) to adapt to parametric uncertainty and unmodeled dynamics in general robotic systems. Our proposed method proceeds by iteratively updating estimates of Lyapunov function derivatives and improving controllers, ultimately yielding a stabilizing quadratic program model-based controller. We validate our approach on a planar Segway simulation, demonstrating substantial performance improvements by iteratively refining on a base model-free controller.

This paper proposes an environmental sound segmentation method using Mask U-Net. Recent research in robot audition has analyzed noise reduction, section detection, and sound source separation for use in a real-world environment with many noises and overlaps. However, conventional methods apply respective functions in cascades. The biggest problem of cascade systems is the accumulation of errors generated at each function block. Although many methods of human voice separation have been proposed, robots operating in a real-world environment must be able to separate not only human voices but other environmental sounds. Unlike traditional sound source separation using spatial information, environmental sound segmentation must simultaneously detect sections and separate sound sources based on pre-trained features. One such method, U-Net, which was proposed for semantic segmentation of images, has been applied to the separation of singing voices. However, this method deals only with limited classes of sounds. The current study proposes an environmental sound segmentation method using Mask U-Net, which combines segmentation using U-Net with sound event detection using CNN to 75-classes of environmental sounds. Experimental application confirmed that this method improved learning speed and sound source separation compared with the conventional method.

With the increasing pace of automation, modern robotic systems need to act in stochastic, non-stationary, partially observable environments. A range of algorithms for finding parameterized policies that optimize for long-term average performance have been proposed in the past. However, the majority of the proposed approaches does not explicitly take into account the variability of the performance metric, which may lead to finding policies that although performing well on average, can perform spectacularly bad in a particular run or over a period of time. To address this shortcoming, we study an approach to policy optimization that explicitly takes into account higher order statistics of the reward function. In this paper, we extend policy gradient methods to include the entropic risk measure in the objective function and evaluate their performance in simulation experiments and on a real-robot task of learning a hitting motion in robot badminton.

This paper presents the design and evaluation of human-like welcoming behaviors for a humanoid robot to draw the attention of passersby by following a three-step model: (1) selecting a target (person) to engage, (2) executing behaviors to draw the target's attention, and (3) monitoring the attentive response. A computer vision algorithm was developed to select the person, start the behaviors and monitor the response automatically. To vary the robot's enthusiasm when engaging passersby, a waving gesture was designed as basic welcoming behavioral element, which could be successively combined with an utterance and an approach movement. This way, three levels of enthusiasm were implemented: Mild (waving), moderate (waving and utterance) and high (waving, utterance and approach movement). The three levels of welcoming behaviors were tested with a Pepper robot at the entrance of a university building. We recorded data and observation sheets from several hundreds of passersby (N = 364) and conducted post-interviews with randomly selected passersby (N = 28). The level selection was done at random for each participant. The passersby indicated that they appreciated the robot at the entrance and clearly recognized its role as a welcoming robot. In addition, the robot proved to draw more attention when showing high enthusiasm (i.e., more welcoming behaviors), particularly for female passersby.

Although imitation learning is often used in robotics, the approach frequently suffers from data mismatch and compounding errors. DAgger is an iterative algorithm that addresses these issues by aggregating training data from both the expert and novice policies, but does not consider the impact of safety. We present a probabilistic extension to DAgger, which attempts to quantity the confidence of the novice policy as a proxy for safety. Our method, EnsembleDAgger, approximates a Gaussian Process using an ensemble of neural networks. Using the variance as a measure of confidence, we compute a decision rule that captures how much we doubt the novice, thus determining when it is safe to allow the novice to act. With this approach, we aim to maximize the novice's share of actions, while constraining the probability of failure. We demonstrate improved safety and learning performance compared to other DAgger variants and classic imitation learning on an inverted pendulum and in the MuJoCo HalfCheetah environment.

In this paper we propose a novel energy-based control scheme for an assist-as-needed rehabilitation strategy, which both adapts the level of support based on patient participation and allows the patient to deviate from the prescribed motion in favor of his/her safety. We build an energy network model, with which we can monitor the energy flow through the system and prescribe a threshold on stored energy. We also develop an adaptive motion control law that shapes the desired trajectory in order to respect the stored energy threshold. Next, we show how adapting the stored energy threshold can be used to change the level of responsiveness to the patient as well as to prevent excessive energy transfer to the human by the system. A criterion is defined for setting this energy threshold, which can be further used for monitoring the patient active participation and for adapting and learning the appropriate assistance level during rehabilitation. Experimental results based on implementation in MATLAB Simscape® and on the VEMO robotic system demonstrate the feasibility of the suggested approach. The presented control scheme can be applied to any system, including position- and torque-controlled robots, and does not require the use of EMG sensors or precise force measurements.

This work introduces a novel 12-motor paired-cable actuation scheme to achieve rolling locomotion with a spherical tensegrity structure. Using a new point mass tensegrity dynamic formulation which we present, we utilize Model Predictive Control to generate optimal state-action trajectories for benchmark evaluation. In particular, locomotive performance is assessed based on the practical criteria of rolling speed, energy efficiency, and directional trajectory-tracking accuracy. Through simulation of 6-motor, 12-motor paired cable, and 24-motor fully-actuated policies, we demonstrate that the 12-motor schema is superior to the 6-motor policy in all benchmark categories, comparable to the 24-motor policy in rolling speed, and is over five times more energy efficient than the fully-actuated 24-motor configuration.

Energy harvesting is an attractive alternative to carrying onboard power for mobile robots, especially for long duration missions. While solar is a powerful option, alternatives are needed for situations where direct sunlight is unavailable. One intriguing concept was proposed in the 17th century to power clocks: energy harvesting based on temporal, rather than spatial, temperature gradients, using a low boiling point fluid that vaporizes at ambient temperatures. This concept has many strengths: it offers all-in-one energy harvesting and storage; direct high-force and large displacement mechanical output, eliminating the need for a motor; and temporal gradients are ubiquitous, due to diurnal thermal fluctuations. The challenge for robotic applications, however, is to create large enough amounts of work in a small enough package to power a mobile device while using a non-toxic and readily available working fluid. Here we present a simple, low-cost energy harvesting actuator, powered by the vaporization of butane and isobutane, with an isobaric energy density of up to 38000 J/m3 (i.e. energy extracted per total volume expansion) each time the temperature fluctuates 13.1°C, enough to power a small car to drive 10m. Two principles enable this: i) precompression of the working fluid, allowing us to tune the boiling point and choose among many non-toxic fluids that do more work than non-compressed fluids; and ii) a constant force profile of the return springs, allowing more work than a linear spring. We present a simple model of the actuator and experimental results characterizing its behavior. Our work lays the foundation for energy harvesting across temporal temperature gradients using vaporization as a viable option for powering mobile robots.

Over the last few years, there has been a significant rise in designing small, agile and flexible medical systems that can navigate through natural orifices. In the case of endoscopic surgery, existing systems vary significantly from each other which raises the question of the existence of a general design that can do it all. In this context, this paper proposes to use a genetic algorithm combined with recorded suturing and anatomical data to automatically design a pair of robotic instruments for the i2 Snake under strict mechanical constraints. The resulting automatically generated instrument designs include a 6 degrees of freedom instrument that can follow a predefined trajectory accurately and a more simple 4 degrees of freedom instrument that can accomplish most of the task. The results also showed the importance of having a prismatic joint to gain the precision required for endoscopic surgery.

This paper studies on sensorimotor control problems of Autonomous Underwater Vehicles (AUVs) using deep reinforcement learning. We design an end-to-end learning architecture mapping original sensor input to continuous control output without referring to the dynamics of vehicles. To avoid difficult and noisy underwater localization, we implement the learning without knowing the positions of AUVs by proposing novel state encoder and reward shaping strategies. Two distinct underwater tasks, obstacle avoidance with sonar sensor and pipeline following with visual sensor, are simulated to validate the effectiveness of proposed architecture and strategies. For the latter, we test the learned policy on realistic images of underwater pipelines to check its generalization ability.