ICRA 2020 Papers

Here we present HAMR-Jr, a 22.5mm, 320mg quadrupedal microrobot. With eight independently actuated degrees of freedom, HAMR-Jr is, to our knowledge, the most mechanically dexterous legged robot at its scale and is capable of high-speed locomotion (13.91bodylengthss-1) at a variety of stride frequencies (1-200Hz) using multiple gaits. We achieved this using a design and fabrication process that is flexible, allowing scaling with minimum changes to our workflow. We further characterized HAMR-Jr's open-loop locomotion and compared it with the larger scale HAMR-VI microrobot to demonstrate the effectiveness of scaling laws in predicting running performance.

Visual topological navigation has been revitalized recently thanks to the advancement of deep learning that substantially improves robot perception. However, the scalability and reliability issue remain challenging due to the complexity and ambiguity of real world images and mechanical constraints of real robots. We present an intuitive approach to show that by accurately measuring the capability of a local controller, large-scale visual topological navigation can be achieved while being scalable and robust. Our approach achieves state-of-the-art results in trajectory following and planning in large-scale environments. It also generalizes well to real robots and new environments without retraining or finetuning.

Autonomous robots often encounter challenging situations where their control policies fail and an expert human operator must briefly intervene, e.g., through teleoperation. In settings where multiple robots act in separate environments, a single human operator can manage a fleet of robots by identifying and teleoperating one robot at any given time. The key challenge is that users have limited attention: as the number of robots increases, users lose the ability to decide which robot requires teleoperation the most. Our goal is to automate this decision, thereby enabling users to supervise more robots than their attention would normally allow for. Our insight is that we can model the user's choice of which robot to control as an approximately optimal decision that maximizes the user's utility function. We learn a model of the user's preferences from observations of the user's choices in easy settings with a few robots, and use it in challenging settings with more robots to automatically identify which robot the user would most likely choose to control, if they were able to evaluate the states of all robots at all times. We run simulation experiments and a user study with twelve participants that show our method can be used to assist users in performing a simulated navigation task. We also run a hardware demonstration that illustrates how our method can be applied to a real-world mobile robot navigation task.

While robot learning has demonstrated promising results for enabling robots to automatically acquire new skills, a critical challenge in deploying learning-based systems is scale: acquiring enough data for the robot to effectively generalize broadly. Imitation learning, in particular, has remained a stable and powerful approach for robot learning, but critically relies on expert operators for data collection. In this work, we target this challenge, aiming to build an imitation learning system that can continuously improve through autonomous data collection, while simultaneously avoiding the explicit use of reinforcement learning, to maintain the stability, simplicity, and scalability of supervised imitation. To accomplish this, we cast the problem of imitation with autonomous improvement into a multi-task setting. We utilize the insight that, in a multi-task setting, a failed attempt at one task might represent a successful attempt at another task. This allows us to leverage the robot's own trials as demonstrations for tasks other than the one that the robot actually attempted. Using an initial dataset of multitask demonstration data, the robot autonomously collects trials which are only sparsely labeled with a binary indication of whether the trial accomplished any useful task or not. We then embed the trials into a learned latent space of tasks, trained using only the initial demonstration dataset, to draw similarities between various trials, enabling the robot to achieve one-shot generalization to new tasks. In contrast to prior imitation learning approaches, our method can autonomously collect data with sparse supervision for continuous improvement, and in contrast to reinforcement learning algorithms, our method can effectively improve from sparse, task-agnostic reward signals.

Gaussian processes (GPs) based Reinforcement Learning (RL) methods with Model Predictive Control (MPC) have demonstrated their excellent sample efficiency. However, since the computational cost of GPs largely depends on the training sample size, learning an accurate dynamics using GPs result in low control frequency in MPC. To alleviate this trade-off and achieve a sample-and-computation-efficient nature, we propose a novel model-based RL method with MPC. Our approach employs a linear Gaussian model with randomized features using the Fastfood as an approximated GP dynamics. Then, we derive an analytic moment-matching scheme in state prediction with the model and uncertain inputs. As a result, the computational cost of the MPC in our RL method does not depend on the training sample size and can improve the control frequency over previous methods. Through experiments with simulated and real robot control tasks, the sample efficiency, as well as the computation efficiency of our model-based RL method, are demonstrated.

Robotic manipulators are reaching a state where we could see them in household environments in the following decade. Nevertheless, such robots need to be easy to instruct by lay people. This is why kinesthetic teaching has become very popular in recent years, in which the robot is taught a motion that is encoded as a parametric function - usually a Movement Primitive (MP)-. This approach produces trajectories that are usually suboptimal, and the robot needs to be able to improve them through trial-and-error. Such optimization is often done with Policy Search (PS) reinforcement learning, using a given reward function. PS algorithms can be classified as model-free, where neither the environment nor the reward function are modelled, or model-based, which can use a surrogate model of the reward function and/or a model for the dynamics of the task. However, MPs can become very high-dimensional in terms of parameters, which constitute the search space, so their optimization often requires too many samples. In this paper, we assume we have a robot motion task characterized with an MP of which we cannot model the dynamics. We build a surrogate model for the reward function, that maps an MP parameter latent space (obtained through a Mutual-information-weighted Gaussian Process Latent Variable Model) into a reward. While we do not model the task dynamics, using mutual information to shrink the task space makes it more consistent with the reward and so the policy improvement is faster in terms of sample efficiency.

We study fundamental theoretical aspects of probabilistic roadmaps (PRM) in the finite time (non-asymptotic) regime. In particular, we investigate how completeness and optimality guarantees of the approach are influenced by the underlying deterministic sampling distribution X and connection radius r > 0. We develop the notion of (δ, ε)-completeness of the parameters X, r, which indicates that for every motion-planning problem of clearance at least δ > 0, PRM using X, r returns a solution no longer than 1+ε times the shortest δ-clear path. Leveraging the concept of e-nets, we characterize in terms of lower and upper bounds the number of samples needed to guarantee (δ, ε)-completeness. This is in contrast with previous work which mostly considered the asymptotic regime in which the number of samples tends to infinity. In practice, we propose a sampling distribution inspired by e-nets that achieves nearly the same coverage as grids while using fewer samples.

Soft robots are theoretically well-suited to rescue and exploration applications where their flexibility allows for the traversal of highly cluttered environments. However, most existing mobile soft robots are not fast or powerful enough to effectively traverse three dimensional environments. In this paper, we introduce a new mobile robot with a continuously deformable slender body structure, the SalamanderBot, which combines the flexibility and maneuverability of soft robots, with the speed and power of traditional mobile robots. It consists of a cable-driven bellows-like origami module based on the Yoshimura crease pattern mounted between sets of powered wheels. The origami structure allows the body to deform as necessary to adapt to complex environments and terrains, while the wheels allow the robot to reach speeds of up to 303.1 mm/s (2.05 body-length/s). Salamanderbot can climb up to 60-degree slopes and perform sharp turns with a minimum turning radius of 79.9 mm (0.54 body-length).

This paper details a novel approach to collision avoidance for aerial vehicles that enables high-speed flight in uncertain environments. This framework is applied at the controller level and provides safety regardless of the planner that is used. The method is shown to be robust to state uncertainty and disturbances, and is computed entirely online utilizing the full nonlinear system dynamics. The effectiveness of this method is shown in a high-fidelity simulation of a quadrotor with onboard sensors rapidly and safely exploring a cave environment utilizing a simple planner.

In many series-elastic actuator applications, the ability to safely render a wide range of impedance is important. Advanced torque control techniques such as the disturbance observer (DOB) can improve torque tracking performance, but their impact on safe impedance range is not established. Here, safety is defined with load port passivity, and passivity conditions are developed for two variants of DOB torque control. These conditions are used to determine the maximum safe stiffness and Z-region of the DOB controllers, which are analyzed and compared with the no DOB case. A feedforward controller is proposed which increases the maximum safe stiffness of the DOB approaches. The results are experimentally validated by manual excitation and in a high-stiffness environment.

This paper considers safe robot mission planning in uncertain dynamical environments. This problem arises in applications such as surveillance, emergency rescue, and autonomous driving. It is a challenging problem due to mod-eling and integrating dynamical uncertainties into a safe planning framework, and finding a solution in a computationally tractable way. In this work, we first develop a probabilistic model for dynamical uncertainties. Then, we provide a framework to generate a path that maximizes safety for complex missions by incorporating the uncertainty model. We also devise a Monte Carlo method to obtain a safe path efficiently. Finally, we evaluate the performance of our approach and compare it to potential alternatives in several case studies.

Autonomous driving in an unregulated urban crowd is an outstanding challenge, especially, in the presence of many aggressive, high-speed traffic participants. This paper presents SUMMIT, a high-fidelity simulator that facilitates the development and testing of crowd-driving algorithms. By leveraging the open-source OpenStreetMap map database and a heterogeneous multi-agent motion prediction model developed in our earlier work, SUMMIT simulates dense, unregulated urban traffic for heterogeneous agents at any worldwide locations that OpenStreetMap supports. SUMMIT is built as an extension of CARLA and inherits from it the physics and visual realism for autonomous driving simulation. SUMMIT supports a wide range of applications, including perception, vehicle control and planning, and end-to-end learning. We provide a context-aware planner together with benchmark scenarios and show that SUMMIT generates complex, realistic traffic behaviors in challenging crowd-driving settings.

Visual place recognition is an important problem in mobile robotics which aims to localize a robot using image information alone. Recent methods have shown promising results for place recognition under varying environmental conditions by exploiting the sequential nature of the image acquisition process. We show that by using k nearest neighbours based image retrieval as the backend, and exploiting the structure of the image acquisition process which introduces temporal relations between images in the database, the location of possible matches can be restricted to a subset of all the images seen so far. In effect, the original problem space can thus be restricted to a significantly smaller subspace, reducing the inference time significantly. This is particularly important for scalable place recognition over databases containing millions of images. We present large scale experiments using publicly sourced data that show the computational performance of the proposed method under varying environmental conditions.

This paper focuses on designing, modeling and validating a novel scalable multicopter whose deformation mechanism, called SNIAE-SSE, relies on a combination of simple non-intersecting angulated elements (SNIAEs) and straight scissor-like elements (SSEs). The proposed SNIAE-SSE mechanism has the advantages of single degree-of-freedom, fast actuation capability and large deformation ratio. In this work, enabled by the SNIAE-SSE mechanism, a quadcopter prototype with symmetrical and synchronous deformation is firstly developed, which facilitates a novel and controllably scalable multicopter system for us to analyze its modeling, as well as to validate its flight performance and dynamics during the deformation in several flight missions including hover, throwing, and morphing flying through a narrow window. Experimental results demonstrate that the developed scalable multicopter can maintain its stable flight behavior even both the folding and unfolding body deformations are fast performed, which indicates an excellent capability of the scalable multicopter to rapidly adapt to complex and dynamically changed environments.

The ability to generate natural language explanations conditioned on the visual perception is a crucial step towards autonomous agents which can explain themselves and communicate with humans. While the research efforts in image and video captioning are giving promising results, this is often done at the expense of the computational requirements of the approaches, limiting their applicability to real contexts. In this paper, we propose a fully-attentive captioning algorithm which can provide state-of-the-art performances on language generation while restricting its computational demands. Our model is inspired by the Transformer model and employs only two Transformer layers in the encoding and decoding stages. Further, it incorporates a novel memory-aware encoding of image regions. Experiments demonstrate that our approach achieves competitive results in terms of caption quality while featuring reduced computational demands. Further, to evaluate its applicability on autonomous agents, we conduct experiments on simulated scenes taken from the perspective of domestic robots.

One of the main challenges of planning legged locomotion in complex environments is the combinatorial contact selection problem. Recent contributions propose to use integer variables to represent which contact surface is selected, and then to rely on modern mixed-integer (MI) optimization solvers to handle this combinatorial issue. To reduce the computational cost of MI, we exploit the sparsity properties of L1 norm minimization techniques to relax the contact planning problem into a feasibility linear program. Our approach accounts for kinematic reachability of the center of mass (COM) and of the contact effectors. We ensure the existence of a quasi-static COM trajectory by restricting our plan to quasi-flat contacts. For planning 10 steps with less than 10 potential contact surfaces for each phase, our approach is 50 to 100 times faster that its MI counterpart, which suggests potential applications for online contact re-planning. The method is demonstrated in simulation with the humanoid robots HRP-2 and Talos over various scenarios.

In this paper we present an approach and a benchmark for visual reasoning in robotics applications, in particular small object grasping and manipulation. The approach and benchmark are focused on inferring object properties from visual and text data. It concerns small household objects with their properties, functionality, natural language descriptions as well as question-answer pairs for visual reasoning queries along with their corresponding scene semantic representations. We also present a method for generating synthetic data which allows to extend the benchmark to other objects or scenes and propose an evaluation protocol that is more challenging than in the existing datasets. We propose a reasoning system based on symbolic program execution. A disentangled representation of the visual and textual inputs is obtained and used to execute symbolic programs that represent a 'reasoning process' of the algorithm. We perform a set of experiments on the proposed benchmark and compare to results from the state of the art methods. These results expose the shortcomings of the existing benchmarks that may lead to misleading conclusions on the actual performance of the visual reasoning systems.

Pose estimation is of paramount importance for flight control as well as localization and navigation of Unmanned Aerial Vehicles (UAVs) to enable autonomous operations. In environments without GPS, such estimation can only be determined using onboard sensors; optical flow using a monocular camera is a popular approach. Monocopters are a class of nature inspired UAVs known as free rotors where their design and flight dynamics are inspired by the falling samara seed. With a constantly rotating body frame, free rotors introduces some unique challenges for visual perception required during optical flow sensing. This paper addresses these problems with the introduction of SHIFT (Selective Heading Image for Translation) that selects optimal images for determining translation with optical flow. It achieves this by decoupling rotation vectors about the optical axis from translation vectors in a flow field through the separate tracking of orientation and position using an Unscented Kalman Filter with phase correlation in the log-polar and spatial domain. The experiments show that SHIFT's estimation in orientation is stable even under sinusoidal excitation with a median absolute percentage errors of less than 1%. It is able to track position and orientation of a UAV accurately.

Learning from observation (LfO) offers a new paradigm for transferring task behavior to robots. LfO requires the robot to observe the task being performed and decompose the sensed streaming data into sequences of state-action pairs, which are then input to LfO methods. Thus, recognizing the state-action pairs correctly and quickly in sensed data is a crucial prerequisite. We present SA-Net a deep neural network architecture that recognizes state-action pairs from RGB-D data streams. SA-Net performs well in two replicated robotic applications of LfO - one involving mobile ground robots and another involving a robotic manipulator - which demonstrates that the architecture could generalize well to differing contexts. Comprehensive evaluations including deployment on a physical robot show that SA-Net significantly improves on the accuracy of the previous methods under various conditions.