IROS 2021 论文列表

With the gradual maturity of the software and hardware of quadruped robots, the application scenarios of quadruped robots are increasing, such as security, rescue, exploration and other tasks. Quadruped robots are flexible and adaptive to challenging or complex environment. This study presents a large-scale quadruped robot, Pegasus II, which is a new version upgraded from the previous quadruped robot, Pegasus [1]. System design of Pegasus II is introduced, including mechanical and electronic design. Locomotion control for a special scene, L-shaped narrow corner, in which a large-scale quadruped robot is not able to traverse in a common quadrupedal mode, is demonstrated. The long body length of a large-scale quadruped robot, such as Pegasus II, incurs difficulty in traversing freely in such a narrow passage. Motivated by this issue, this study proposes an experimental implementation to realize the transition from quadrupedal mode to bipedal mode. The control framework is presented, which mainly includes trajectory optimization, whole-body control, compliance control, and joint torque estimator. Simulations and experiments are conducted to validate the performance, including gait transition, compliance control.

We report the results of our study on whether taking a walk with a child-like robot that a person takes care of can generate interaction with the surrounding people whom the person has never met before. As the number of single-person households increases, it is expected that more people will live with not only robots that can be useful for people but also robots that people can take care of. Our study is important because we explored the possibilities of such a robot to create interaction between people in the community, which has been lost in recent years. In this paper, we designed the behavior of a robot that follows a person and learns about the scenery while walking together and implemented it using Pepper. Then, the first author walked around the university building with the robot and observed the initial reactions of the surrounding people. As we expected, the surrounding people interacted with the first author by talking to the robot as if it were a child and helping the robot. This paper contributes to taking a step forward a new research theme ‘taking a walk with a robot and a person’ by focusing on the important role of robots in the future relationship with a person: care-receiving from a person and communication mediation between a person and others.

This paper presents a stumble recovery controller for a knee exoskeleton that detects a stumble perturbation; selects an anticipated recovery strategy; and provides appropriate recovery assistance. In order to assess the efficacy of the controller in providing an assistive response to a stumble perturbation, the controller was implemented in a knee exoskeleton and evaluated in a single healthy adult participant against several other controller reactions, and against the participant’s response without an exoskeleton. Results show that the stumble recovery controller successfully detected the perturbation and correctly selected the strategy that matched the participant’s response for all 29 trials in which the exoskeleton was used. Further, results show improvements in stumble recovery metrics when using the exoskeleton with the stumble recovery controller, compared to the control cases of: 1) no change in the nominal controller when stumble is detected; 2) turning off exoskeleton torque when a stumble is detected; and 3) not wearing an exoskeleton.

Flapping-wing robots (so-called ornithopters) are a promising type of platform to perform efficient winged flight and interaction with the environment. However, the control of such vehicles is challenging due to their under-actuated morphology to meet lightweight requirements. Consequently, the flight control of flapping-wing robots is predominantly handled by the tail. Most ornithopters feature a tail with two degrees of freedom but the configuration choice is often arbitrary and without in-depth study. In this paper, we propose a thorough analysis of the design and in-flight performance for three tails. Their design and manufacturing methods are presented, with an emphasis on low weight, which is critical in ornithopters. The aerodynamics of the tails is analyzed through CFD simulations and their performance compared experimentally. The advantages and performance metrics of each configuration are discussed based on flight data. Two types of 3D flight tests were carried out: aggressive heading maneuvers and level turns. The results show that an inverted V-tail outperforms the others regarding maneuverability and stability. From the three configurations, only the inverted V-Tail can perform an aggressive stable banked level turn with a radius of 3.7 m at a turning rate of 1.6 rad/s. This research work describes the impact of the tail configuration choice on the performance of bird-scale flapping-wing robots.

To cope with the requirements on efficiency and labour-saving of the out-pipe surface grinding tasks in the wild, several proposals are revealed and discussed. The one benefiting from the characteristics of planetary gear transmission and friction actuating mechanism are expatiated. To realize full coverage of out-pipe surface, the self-rotation and revolution motions of every polishing tool (cutter) are actuated by the same motor, and the friction force produced in grinding process acts as suitable tractive force for the forward travel of the grinding system. The friction statics analysis is established to illustrate the force transmission. Compression spring system are utilized to realize force equilibrium and support passive diameter adaptability. The proposed robotic grinding system is characterized by less actuator, online grinding capability and high working efficiency. It has clear advantages regarding manufacturing costs and control complexity. As the result of prototype experiments, performance of smooth grinding the out-pipe surface is confirmed.

Here we developed an experimental platform with a magnetic, modular, undulatory robot (μBot) for studying fish-inspired underwater locomotion. This platform will enable us to systematically explore the relationship between body morphology, swimming gaits, and swimming performance via reinforcement learning methods. The μBot was designed to be easily modifiable in morphology, compact in size, easy to be controlled and inexpensive. The experimental platform also included a towing tank and a motion tracking system for real-time measurement of the μBot kinematics. The swimming gaits of μBot were generated by a central pattern generator (CPG), which outputs voltage signals to μBot's magnetic actuators. The CPG parameters were learned experimentally using the parameter exploring policy gradient (PGPE) method to maximize swimming speed. In the experiments, two μBot designs with the same body morphology but different caudal-fin shapes were tested. Results showed that swimming gaits with back-propagating traveling waves can be learned experimentally via PGPE, while the shape of the caudal fins had moderate influences on the learned gaits and the swimming speed. Furthermore, robot swimming speed was sensitive to the undulating frequency and the voltage magnitude of the last three posterior actuators. In contrast, swimming gaits and speed were relatively invariant to the variances within the inter-module connection weights of CPG and the voltage applied to the anterior actuator.

This work introduces an approach for automatic hair combing by a lightweight robot. For people living with limited mobility, dexterity, or chronic fatigue, combing hair is often a difficult task that negatively impacts personal routines. We propose a modular system for enabling general robot manipulators to assist with a hair-combing task. The system consists of three main components. The first component is the segmentation module, which segments the location of hair in space. The second component is the path planning module that proposes automatically-generated paths through hair based on user input. The final component creates a trajectory for the robot to execute. We quantitatively evaluate the effectiveness of the paths planned by the system with 48 users and qualitatively evaluate the system with 30 users watching videos of the robot performing a hair-combing task in the physical world. The system is shown to effectively comb different hairstyles.

This article presents a novel transformable wheeled robot with three motion modes. Based on the four-bar mechanism design, the transformable wheel can transform into CW (clockwise) legged wheel mode and CCW (counterclock-wise) legged wheel mode from the circular wheel. Both legged wheel modes achieve good obstacle-climbing performance when the robot overcomes obstacles such as steps in the forward and backward directions. Mathematical expression of shape of the transformable wheel is established for design variables selection of the wheel to improve the obstacle-overcoming ability of the legged wheel modes. Dynamics model of the robot in the step-overcoming scene is analyzed to determine and reduce the torque load of the motor. Finally, two prototypes of the robot are developed. The early-stage prototype made by 3D printing verifies the step-overcoming ability of the transformable wheel. Then, structural strength of the robot chassis and the robustness of the hardware are improved in the second prototype, which can play a role in SAR and indoor service tasks. In general, the prototype can overcome steps with the height of more than 2 times the radius of the wheel in CW/CCW legged wheel mode and move at a maximum speed of 3.15 m/s in circular wheel mode (about 5.83 times the length of the robot per second).

Musculoskeletal humanoids have various biomimetic advantages, and the redundant muscle arrangement allowing for variable stiffness control is one of the most important. In this study, we focus on one feature of the redundancy, which enables the humanoid to keep moving even if one of its muscles breaks, an advantage that has not been dealt with in many studies. In order to make the most of this advantage, the design of muscle arrangement is optimized by considering the maximization of minimum available torque that can be exerted when one muscle breaks. This method is applied to the elbow of a musculoskeletal humanoid Musashi with simulations, the design policy is extracted from the optimization results, and its effectiveness is confirmed with the actual robot.

Transparent objects with unique visual properties often make depth cameras fail to scan their reflective and refractive surfaces. Recent studies on depth completion of transparent objects have leveraged a linear system based on the geometric constraints to predict the missing depth, which is hard to be employed in an end-to-end framework and achieve joint optimization. In this paper, we propose DepthGrasp - a deep learning approach for depth completion of transparent objects from a raw RGB-D image. More specifically, we use a generative adversarial network, which utilizes the generator to complete the depth maps by predicting the missing or inaccurate depth values, and use discriminator to guide the completed depth maps against the groundtruth. In the generator, we devise spectral residual blocks (SRB) with spectral normalization for network stability, and residual block to pass the attention map in order to capture the structure information and distinguish the geometric shape of transparent objects. In the discriminator, we use a patch-based convolutional network to adapt the data distributions of the predicted depth maps according to groundtruth. Extensive experiments conducted on ClearGrasp dataset show the effectiveness and generalization of the DepthGrasp for depth completion, and the deployed robotic picking system makes significant improvement on the performance of grasping on transparent objects.

RGB-D cameras have been successfully used for indoor High-ThroughPut Phenotyping (HTPP). However, their capability and feasibility for in-field HTPP applications still need to be evaluated. To solve the problem, we evaluate the depth-ranging performances of a consumer-level RGB-D camera (RealSense D435i) under in-field scenarios. First, we focus on determining their optimal ranging areas for different crop organs. Second, based on the evaluation results, we analyze the influences of light intensity on depth measurements and propose a brightness-and-distance based Support Vector Regression Strategy, to compensate the ranging error. Finally, we give an intuitive accuracy ranking diagram for RealSense D435i under natural lighting intensities. Experimental results show that: 1) RealSense D435i has good ranging performances on in-field HTPP. 2) Our error compensation model can effectively reduce the influences of lighting intensity and target distance.

We tackle the problem of refining and denoising a series of 3D human poses estimated from a low-resolution video. Low-resolution often causes the wrong pose estimation, e.g., left-right switching and the absence of keypoints. We propose to use the variational autoencoder (VAE) to remove these challenging noises. The VAE model utilizes time-series information and motion priors in denoising. From our experiments, the VAE model can reduce the pose estimation error (MPJPE) for poor-quality images by 24.37mm, from the original 105.53mm. This improves about 6.5 times over the traditional DCT approach. In addition, it removes jitters and generates smooth movements, which is helpful in recognition of human behaviors.

The ability to successfully grasp objects is crucial in robotics, as it enables several interactive downstream applications. To this end, most approaches either compute the full 6D pose for the object of interest or learn to predict a set of grasping points. While the former approaches do not scale well to multiple object instances or classes yet, the latter require large annotated datasets and are hampered by their poor generalization capabilities to new geometries. To overcome these shortcomings, we propose to teach a robot how to grasp an object with a simple and short human demonstration. Hence, our approach neither requires many annotated images nor is it restricted to a specific geometry. We first present a small sequence of RGB-D images displaying a human-object interaction. This sequence is then leveraged to build associated hand and object meshes that represent the depicted interaction. Subsequently, we complete missing parts of the reconstructed object shape and estimate the relative transformation between the reconstruction and the visible object in the scene. Finally, we transfer the a-priori knowledge from the relative pose between object and human hand with the estimate of the current object pose in the scene into necessary grasping instructions for the robot. Exhaustive evaluations with Toyota’s Human Support Robot (HSR) in real and synthetic environments demonstrate the applicability of our proposed methodology and its advantage in comparison to previous approaches.

General purpose simulators provide cheap training data to learn complex robotic skills. However, the transition from simulation to reality is often very challenging for the agent. One major issue is the delay on the physical robot that may deteriorate the performance of the deployed agent. Furthermore, once a successfully trained learning-based control policy is available, re-purposing the knowledge acquired by the agent to enable a structurally distinct agent to perform the same task is hazardous if done naively. In this work, we address the above issues with a single method, the DA-UNN (Delay Aware Universal Notice Network), which decomposes the knowledge into robot-specific and task-specific modules for fast transfer. Our framework deals with delays immanent to physical systems in order to improve sim2real transfer. We evaluate the efficiency of our approach using simulated and actual robots on a dynamic manipulation task where delay management is crucial.

The paper focuses on collision-inclusive motion planning for impact-resilient mobile robots. We propose a new deformation recovery and replanning strategy to handle collisions that may occur at run-time. Contrary to collision avoidance methods that generate trajectories only in conservative local space or require collision checking that has high computational cost, our method directly generates (local) trajectories with imposing only waypoint constraints. If a collision occurs, our method then estimates the post-impact state and computes from there an intermediate waypoint to recover from the collision. To achieve so, we develop two novel components: 1) a deformation recovery controller that optimizes the robot’s states during post-impact recovery phase, and 2) a post-impact trajectory replanner that adjusts the next waypoint with the information from the collision for the robot to pass through and generates a polynomial-based minimum effort trajectory. The proposed strategy is evaluated experimentally with an omnidirectional impact-resilient wheeled robot. The robot is designed in house, and it can perceive collisions with the aid of Hall effect sensors embodied between the robot’s main chassis and a surrounding deflection ring-like structure.

In this paper, we presented a new method for deformation control of deformable objects, which utilizes both visual and tactile feedback. At present, manipulation of deformable objects is basically formulated by assuming positional constraints. But in fact, in many situations manipulation has to be performed under actively applied force constraints. This scenario is considered in this research. In the proposed scheme a tactile feedback is integrated to ensure a stable contact between the robot end-effector and the soft object to be manipulated. The controlled contact force is also utilized to regulate the deformation of the soft object with its shape measured by a vision sensor. The effectiveness of the proposed method is demonstrated by a book page turning and shaping experiment.

Deformable solid objects such as clay or dough are prevalent in industrial and home environments. However, robotic manipulation of such objects has largely remained unexplored in literature due to the high complexity involved in representing and modeling their deformation. This work addresses the problem of shaping elasto-plastic dough by proposing to use a novel elastic end-effector to roll dough in a reinforcement learning framework. The transition model for the end-effector-to-dough interactions is learned from one hour of robot exploration, and doughs of different hydration levels are rolled out into varying lengths. Experimental results are encouraging, with the proposed framework accomplishing the task of rolling out dough into a specified length with 60% fewer actions than a heuristic method. Furthermore, we show that estimating stiffness using the soft end-effector can be used to effectively initialize models, improving robot performance by approximately 40% over incorrect model initialization.

Testing and verification is still an open issue on the way to fully automated driving. Simulations can help to reduce the required testing efforts, however, classical simulators based on physical models and heuristics, such as the intelligent driver model (IDM), show limited model accuracy on a microscopic scenario level. In turn, learning-based driver models are often capable to predict human driver’s behavior accurately, but are difficult to tailor such that they follow an intended scenario description. In this work, we propose a software-in-the-loop framework to combine a learned model with a rule-based logic layer and a kinematic vehicle model of a classical traffic simulator. Thus, the merits of both, classical simulators and learning-based models are exploited. We demonstrate with a case study of evaluating a motion planning scheme that the simulator fits well with the needs of testing such methods. Furthermore, we show by experiments with real-world traffic data from a traffic surveillance system that the proposed simulator yields realistic behavior of the simulated road users.

Traditional monocular visual simultaneous localization and mapping (SLAM) algorithms have been extensively studied and proven to reliably recover a sparse structure and camera motion. Nevertheless, the sparse structure is still insufficient for scene interaction, e.g., visual navigation and augmented reality applications. To densify the scene reconstruction, the use of single-image absolute depth prediction from convolutional neural networks (CNNs) for filling in the missing structure has been proposed. However, the prediction accuracy tends to not generalize well on scenes that are different from the training datasets.In this paper, we propose a dense monocular SLAM system, named DeepRelativeFusion, that is capable to recover a globally consistent 3D structure. To this end, we use a visual SLAM algorithm to reliably recover the camera poses and semi-dense depth maps of the keyframes, and then use relative depth prediction to densify the semi-dense depth maps and refine the keyframe pose-graph. To improve the semi-dense depth maps, we propose an adaptive filtering scheme, which is a structure- preserving weighted average smoothing filter that takes into account the pixel intensity and depth of the neighbouring pixels, yielding substantial reconstruction accuracy gain in densification. To perform densification, we introduce two incremental improvements upon the energy minimization framework proposed by DeepFusion: (1) an improved cost function, and(2) the use of single-image relative depth prediction. After densification, we update the keyframes with two-view consistent optimized semi-dense and dense depth maps to improve pose- graph optimization, providing a feedback loop to refine the keyframe poses for accurate scene reconstruction. Our system outperforms the state-of-the-art dense SLAM systems quantitatively in dense reconstruction accuracy by a large margin.For more information, see the demo video and supplementary material.