ICRA 2022 Papers

Autonomous navigation of mobile robots is an es-sential aspect in use cases such as delivery, assistance or logistics. Although traditional planning methods are well integrated into existing navigation systems, they struggle in highly dynamic en-vironments. On the other hand, Deep-Reinforcement-Learning-based methods show superior performance in dynamic obstacle avoidance but are not suitable for long-range navigation and struggle with local minima. In this paper, we propose a Deep-Reinforcement-Learning-based control switch, which has the ability to select between different planning paradigms based solely on sensor data observations. Therefore, we develop an interface to efficiently operate multiple model-based, as well as learning-based local planners and integrate a variety of state-of-the-art planners to be selected by the control switch. Subsequently, we evaluate our approach against each planner individually and found improvements in navigation performance especially for highly dynamic scenarios. Our planner was able to prefer learning-based approaches in situations with a high number of obstacles while relying on the traditional model-based planners in long corridors or empty spaces.

Loop closure detection is an important building block that ensures the accuracy and robustness of simultaneous localization and mapping (SLAM) systems. Due to their generalization ability, CNN-based approaches have received increasing attention. Although they normally benefit from training on datasets that are diverse and reflective of the environments, new environments often emerge after the model is deployed. It is therefore desirable to incorporate the data newly collected during operation for incremental learning. Nevertheless, simply finetuning the model on new data is infeasible since it may cause the model's performance on previously learned data to degrade over time, which is also known as the problem of catastrophic forgetting. In this paper, we present AirLoop, a method that leverages techniques from lifelong learning to minimize forgetting when training loop closure detection models incrementally. We experimentally demonstrate the effectiveness of AirLoop on TartanAir, Nordland, and RobotCar datasets. To the best of our knowledge, AirLoop is one of the first works to achieve lifelong learning of deep loop closure detectors.

Dynamic Object-aware SLAM (DOS) exploits object-level information to enable robust motion estimation in dynamic environments. Existing methods mainly focus on identifying and excluding dynamic objects from the optimization. In this paper, we show that feature-based visual SLAM systems can also benefit from the presence of dynamic articulated objects by taking advantage of two observations: (1) The 3D structure of each rigid part of articulated object remains consistent over time; (2) The points on the same rigid part follow the same motion. In particular, we present AirDOS, a dynamic object-aware system that introduces rigidity and motion constraints to model articulated objects. By jointly optimizing the camera pose, object motion, and the object 3D structure, we can rectify the camera pose estimation, preventing tracking loss, and generate 4D spatio-temporal maps for both dynamic objects and static scenes. Experiments show that our algorithm improves the robustness of visual SLAM algorithms in challenging crowded urban environments. To the best of our knowledge, AirDOS is the first dynamic object-aware SLAM system demonstrating that camera pose estimation can be improved by incorporating dynamic articulated objects.

Robots operating in human-centered environments should have the ability to understand how objects function: what can be done with each object, where this interaction may occur, and how the object is used to achieve a goal. To this end, we propose a novel approach that extracts a self-supervised visual affordance model from human teleoperated play data and leverages it to enable efficient policy learning and motion planning. We combine model-based planning with model-free deep reinforcement learning (RL) to learn policies that favor the same object regions favored by people, while requiring minimal robot interactions with the environment. We evaluate our algorithm, Visual Affordance-guided Policy Optimization (VAPO), with both diverse simulation manipulation tasks and real world robot tidy-up experiments to demonstrate the effectiveness of our affordance-guided policies. We find that our policies train 4 × faster than the baselines and generalize better to novel objects because our visual affordance model can anticipate their affordance regions.

Collaborative heterogeneous robots are expected to perform comprehensive perception, mapping and coordination in search and rescue scenarios. The challenge of collaboration between heterogeneous robots lies in their huge differences in perception, mobility and processing capabilities. In this paper, a novel collaborative UAV-UGV mapping framework is proposed in GNSS-denied and unknown environments. The key novelty of this work is the proposing of a unified framework to formulate the UAV-UGV collaborative mapping problem with a continuous-discrete model, as well as its realization in real robotic systems. In order to project continuous space into discrete space, a novel information gain trigger scheme is pro-posed. The continuous space allows each robot to perform high frequency local map estimation, while discrete space describes the problem of multi-resolution hybrid map fusion. Considering the nature of data heterogeneity, a flexible probabilistic fusion algorithm is proposed that addresses the multi-resolution hybrid map fusion problem, where the local maps generated by UAV and UGV are fused based on Bayesian rule. The proposed UAV-UGV hybrid system is validated in various challenging scenarios, demonstrating its accuracy and utility in practical tasks.

In recent years, an increasing number of research works have been focusing on the manipulation by aerial robots. Previous works using aerial robots with robotic arms have two problems: underactuation and external disturbances. We propose the fully-actuated control method and motion strategy using contact with the environment to solve these problems, along with the mechanical approach required. First, each propeller's 1 degree-of-freedom (DoF) thrust vectoring units are applied to enable fully-actuated flight control. In order to obtain the desired thrust and vectoring angle inputs for aerial manipulation satisfying hardware limits, we developed a fully-actuated control method using non-linear optimization. Second, we propose a manipulation motion strategy that treats the multilink robot body as a fixed manipulator by making contact with the environment. The contact mechanism attached to the link end is developed to maintain contact and resist external disturbances. In a real machine experiment, the robot successfully opened the door while in contact with the wall, demonstrating the feasibility of the proposed methods.

The imitation learning research community has recently made significant progress towards the goal of enabling artificial agents to imitate behaviors from video demonstrations alone. However, current state-of-the-art approaches developed for this problem exhibit high sample complexity due, in part, to the high-dimensional nature of video observations. Towards addressing this issue, we introduce here a new algorithm called Visual Generative Adversarial Imitation from Observation using a State Observer (VGAIfO-SO). At its core, VGAIfO-SO seeks to address sample inefficiency using a novel, self-supervised state observer, which provides estimates of lower-dimensional proprioceptive state representations from high-dimensional images. We show experimentally in several continuous control environments that VGAIfO-SO is more sample efficient than other IfO algorithms at learning from video-only demonstrations and can sometimes even achieve performance close to the Generative Adversarial Imitation from Observation (GAIfO) algorithm that has privileged access to the demonstrator's proprioceptive state information.

This paper presents an optimization of an admittance control model for gait balance assistance offered by a walker-type assistive robot. We previously introduced the notion of quasi-passive physical Human-Robot Interaction (pHRI) where a non-wearable assistive device adaptively achieves supportability for providing physical assistance and operability to follow the user's intuitive operation. Aiming to mitigate the falling risk of elderly people with reduced mobility with our pHRI approach, we propose a hierarchical algorithm to optimize an admittance control model for a walker robot. By employing dynamic trajectories such as Zero Moment Point (ZMP) and Divergent Component of Motion (DCM) with optimization, our controller provides appropriate physical interaction to improve the gait stability while considering intrinsic body dynamics. In the current implementation, based on a model predictive control (MPC) framework, we formulate the optimization problems in the form of quadratic programming (QP), making the optimization suitable for real-time interaction. Through mechanical assessments with passive walking models of compass gait, we demonstrate the feasibility of our proposed optimization framework in stabilizing the limit cycle gait with minimized assistance.

Human-in-the-loop (HIL) optimization usually optimizes assistive torque of exoskeletons to minimize the human's energetic expenditure in walking, quantified by metabolic cost. This formulation can, however, result in altered gait pattern of the human joint from the natural pattern, which is undesired. In this paper, we proposed a novel concept of HIL optimization of a hip exoskeleton. The optimization goal was to maintain the hip kinematics while providing optimal mechanical energy from the exoskeleton by modulating the admittance control. Policy iteration was used to optimize the switching time within the gait phase, at which a single parameter of the admittance controller was altered to provide assistance. The stiffness and equilibrium angle were considered as the two parameters for altering at the switching time, resulting in three possible modes of operation for the algorithm: (i) switching the equilibrium point, (ii) switching stiffness while equilibrium point is set at maximum extension and, (iii) maximum flexion. The optimization algorithm was found to converge for all three modes, with the equilibrium mode resulting in multiple solutions. Further analysis of power injected by the exoskeleton in the three modes showed that the first and third mode reduced human energetic exertion while the second mode increased human exertion. Implications of the results as well as the observed muscle activation patterns in response to assistance are discussed.

Human-robot collaboration aims to extend human ability through cooperation with robots. This technology is currently helping people with physical disabilities, has transformed the manufacturing process of companies, improved surgical performance, and will likely revolutionize the daily lives of everyone in the future. Being able to enhance the performance of both sides, such that human-robot collaboration outperforms a single robot/human, remains an open issue. For safer and more effective collaboration, a new control scheme has been proposed for redundant robots in this paper, consisting of an adaptive vision-based control term in task space and an interactive control term in null space. Such a formulation allows the robot to autonomously carry out tasks in an unknown environment without prior calibration while also interacting with humans to deal with unforeseen changes (e.g., potential collision, temporary needs) under the redundant configuration. The decoupling between task space and null space helps to explore the collaboration safely and effectively without affecting the main task of the robot end-effector. The stability of the closed-loop system has been rigorously proved with Lyapunov methods, and both the convergence of the position error in task space and that of the damping model in null space are guaranteed. The experimental results of a robot manipulator guided with the technology of augmented reality (AR) are presented to illustrate the performance of the control scheme.

The task space control of robot manipulators requires solving the thorny problem of stabilizing the compound system {outer controller - inner controller - robot manipulator}. To stabilize this compound system, both controllers must be designed by the user to achieve convergence of the tracking error. This problem is tricky to solve in the case of the control of an industrial robot manipulator because its internal controller is not accessible to users. In this work, we propose an adaptive neural network outer controller. The neural networks approximate the dynamics of the inner controller, the kinematic and dynamic parameters of the robot. Besides, the adaptive part finds parameters that achieve the stability of the global system. Since an adaptive approach is sensitive to errors in initial values, we have integrated into the controller a term that constrains the closed-loop system to maintain the prescribed performances. The effectiveness of the approach is demonstrated through Lyapunov's theory, simulation comparisons, and experimental studies.

In order to provide therapy in a functional context, controls for wearable robotic orthoses need to be robust and intuitive. We have previously introduced an intuitive, user-driven, EMG-based method to operate a robotic hand orthosis, but the process of training a control that is robust to concept drift (changes in the input signal) places a substantial burden on the user. In this paper, we explore semi-supervised learning as a paradigm for controlling a powered hand orthosis for stroke subjects. To the best of our knowledge, this is the first use of semi-supervised learning for an orthotic application. Specifically, we propose a disagreement-based semi-supervision algorithm for handling intrasession concept drift based on multimodal ipsilateral sensing. We evaluate the performance of our algorithm on data collected from five stroke subjects. Our results show that the proposed algorithm helps the device adapt to intrasession drift using unlabeled data and reduces the training burden placed on the user. We also validate the feasibility of our proposed algorithm with a functional task; in these experiments, two subjects successfully completed multiple instances of a pick-and-handover task.

Aerial robots are increasingly being utilized for environmental monitoring and exploration. However, a key challenge is efficiently planning paths to maximize the information value of acquired data as an initially unknown environment is explored. To address this, we propose a new approach for informative path planning based on deep reinforcement learning (RL). Combining recent advances in RL and robotic applications, our method combines tree search with an offline-learned neural network predicting informative sensing actions. We introduce several components making our approach applicable for robotic tasks with high-dimensional state and large action spaces. By deploying the trained network during a mission, our method enables sample-efficient online replanning on platforms with limited computational resources. Simulations show that our approach performs on par with existing methods while reducing runtime by 8-10×. We validate its performance using real-world surface temperature data.

The problem addressed in this work is the arbitration of the role between a robot and a human during physical Human-Robot Interaction, sharing a common task. The system is modeled as a Cartesian impedance, with two separate external forces provided by the human and the robot. The problem is then reformulated as a Cooperative Differential Game, which possibly has multiple solutions on the Pareto frontier. Finally, the bargaining problem is addressed by proposing a solution depending on the interaction force, interpreted as the human will to lead or follow. This defines the arbitration law and assigns the role of leader or follower to the robot. Experiments show the feasibility and capabilities of the proposed control in managing the human-robot arbitration during a shared- trajectory following task.

Soft robots are made of compliant materials and perform tasks that are challenging for rigid robots. However, their continuum nature makes it difficult to develop model-based control strategies. This work presents a robust model-based control scheme for soft continuum robots. Our dynamic model is based on the Euler-Lagrange approach, but it uses a more accurate description of the robot's inertia and does not include oversimplified assumptions. Based on this model, we introduce an adaptive sliding mode control scheme, which is robust against model parameter uncertainties and unknown input disturbances. We perform a series of experiments with a physical soft continuum arm to evaluate the effectiveness of our controller at tracking task-space trajectory under different payloads. The tracking performance of the controller is around 38 % more accurate than that of a state-of-the-art controller, i.e., the inverse dynamics method. Moreover, the proposed model-based control design is flexible and can be generalized to any continuum robotic arm with an arbitrary number of segments. With this control strategy, soft robotic object manipulation can become more accurate while remaining robust to disturbances.

Vital signs such as heart rate, oxygen saturation, and blood pressure are crucial information for healthcare workers to identify clinical deterioration of ward patients. Currently, medical devices monitor these vital signs and trigger alarms when the vital signs are not in the normal ranges based on predefined thresholds, which suggests the presence of clinical deterioration. However, such threshold-based approach is not robust for patient monitoring. This is because vital signs differ among patients due to human physiology and change across time based on the action performed by a patient. In this work, we want to tackle these problems by building adaptable action-aware vital models. These models can understand the changes in vital signs caused by patient's actions and can be adapted to the normal vital sign ranges of individual patients. Our experimental results show that general vital sign patterns for different actions exist and can be personalized to new patients. Additionally, we investigate the possibility of estimating the initial vital model for an unobserved action using models of observed actions for model personalization. The resulting adaptable action-aware vital models have the potential to improve patient monitoring by reducing false clinical alarms.

In this paper, we propose an efficient frontier detector method based on adaptive Rapidly-exploring Random Tree (RRT) for autonomous robot exploration. Robots can achieve real-time incremental frontier detection when they are exploring unknown environments. First, our detector adaptively adjusts the sampling space of RRT by sensing the surrounding environment structure. The adaptive sampling space can greatly improve the successful sampling rate of RRT (the ratio of the number of samples successfully added to the RRT tree to the number of sampling attempts) according to the environment structure and control the expansion bias of the RRT. Second, by generating non-uniform distributed samples, our method also solves the over-sampling problem of RRT in the sliding windows, where uniform random sampling causes over-sampling in the overlap area between two adjacent sliding windows. In this way, our detector is more inclined to sample in the latest explored area, which improves the efficiency of frontier detection and achieves incremental detection. We validated our method in three simulated benchmark scenarios. The experimental comparison shows that we reduce the frontier detection runtime by about 40% compared with the SOTA method, DSV Planner.

Visual tracking is adopted to extensive unmanned aerial vehicle (UAV)-related applications, which leads to a highly demanding requirement on the robustness of UAV trackers. However, adding imperceptible perturbations can easily fool the tracker and cause tracking failures. This risk is often overlooked and rarely researched at present. Therefore, to help increase awareness of the potential risk and the robustness of UAV tracking, this work proposes a novel adaptive adversarial attack approach, i.e., Ad2 Attack, against UAV object tracking. Specifically, adversarial examples are generated online during the resampling of the search patch image, which leads trackers to lose the target in the following frames. Ad2 Attack is composed of a direct downsampling module and a super-resolution upsampling module with adaptive stages. A novel optimization function is proposed for balancing the imperceptibility and efficiency of the attack. Comprehensive experiments on several well-known benchmarks and real-world conditions show the effectiveness of our attack method, which dramatically reduces the performance of the most advanced Siamese trackers.

Testing and evaluation of field robotic systems requires both experimentation in representative conditions and human supervision to effectively assess components, manage risk, and interpret results. Due to the complexity of robotic sys-tems, we argue this experimentation should be done adaptively by using insights gained from previous trials. Furthermore, we envision an advisory system that could assist experimenters with selecting trial configurations by learning and accounting for human preferences and risk tolerances; however, formal methods for human decision making in the context of field robotic experimentation remains an open question. In this work, we present and analyze a case study for how decisions were made during the testing and evaluation of an off-road, autonomous navigation system. From the perspective of active learning, we find that Bayesian Optimization is a promising mathematical framework for modeling human decision making in adaptive experimental design of field robotics and that a combination of the EI, KG, and PES acquisition functions would likely be useful for realizing an advisory system.