ICRA 2023 论文列表

Perceptual understanding of the scene and the relationship between its different components is important for successful completion of robotic tasks. Representation learning has been shown to be a powerful technique for this, but most of the current methodologies learn task specific representations that do not necessarily transfer well to other tasks. Furthermore, representations learned by supervised methods require large, labeled datasets for each task that are expensive to collect in the real-world. Using self-supervised learning to obtain representations from unlabeled data can mitigate this problem. However, current self-supervised representation learning methods are mostly object agnostic, and we demonstrate that the resulting representations are insufficient for general purpose robotics tasks as they fail to capture the complexity of scenes with many components. In this paper, we show the effectiveness of using object-aware representation learning techniques for robotic tasks. Our self-supervised representations are learned by observing the agent freely interacting with different parts of the environment and are queried in two different settings: (i) policy learning and (ii) object location prediction. We show that our model learns control policies in a sample-efficient manner and outperforms state-of-the-art object agnostic techniques as well as methods trained on raw RGB images. Our results show a 20% increase in performance in low data regimes (1000 trajectories) in policy training using implicit behavioral cloning (IBC). Furthermore, our method outperforms the baselines for the task of object localization in multi-object scenes. Further qualitative results are available at https://sites.google.com/view/slots4robots.

Implementing dynamic locomotion behaviors on legged robots requires a high-quality state estimation module. Especially when the motion includes flight phases, state-of-the-art approaches fail to produce reliable estimation of the robot posture, in particular base height. In this paper, we propose a novel approach for combining visual-inertial odometry (VIO) with leg odometry in an extended Kalman filter (EKF) based state estimator. The VIO module uses a stereo camera and IMU to yield low-drift 3D position and yaw orientation and drift-free pitch and roll orientation of the robot base link in the inertial frame. However, these values have a considerable amount of latency due to image processing and optimization, while the rate of update is quite low which is not suitable for low-level control. To reduce the latency, we predict the VIO state estimate at the rate of the IMU measurements of the VIO sensor. The EKF module uses the base pose and linear velocity predicted by VIO, fuses them further with a second high-rate IMU and leg odometry measurements, and produces robot state estimates with a high frequency and small latency suitable for control. We integrate this lightweight estimation framework with a nonlinear model predictive controller and show successful implementation of a set of agile locomotion behaviors, including trotting and jumping at varying horizontal speeds, on a torque-controlled quadruped robot.

Ultrasound (US) is widely used in image-guided needle procedures. Correctly tracking the needle tip position in US images during the procedure plays an important role in improving the needle targeting accuracy and patient safety. This paper presents a leaning-based visual tracking network with a Siamese architecture, which makes full use of the attention mechanism to explore the potential of global features and takes advantage of an online target model prediction module to robustly track the needle tip in US images. Several self- and cross-attention modules are applied to learn global features from the whole US image. A discriminative target model is also learned as a complementary part to improve the discriminability of the proposed tracker. The template used during the tracking is updated frequently according to the tracking results to ensure that the tracker can always capture the latest characteristics of the appearance of the needle tip. Experimental results in both phantom and tissue showed that the proposed tracking network was more robust than other state-of-the-art visual trackers. The mean success rates of the proposed tracker are 7.1% and 9.2% higher than the second best performing visual tacker when the needle was inserted by motors and human hands in the tissue experiments.

Motion estimation is an essential element for autonomous vessels. It is used e.g. for lidar motion compensation as well as mapping and detection tasks in a maritime environment. Because the use of gyroscopes is not reliable and a high performance inertial measurement unit is quite expensive, we present an approach for visual pitch and roll estimation that utilizes a convolutional neural network for water segmentation, a stereo system for reconstruction and simple geometry to estimate pitch and roll. The algorithm is validated on a novel, publicly available dataset22https://git.ios.htwg-konstanz.de/dgriesse/constance_orientation_dataset/archive/main/constance_orientation_dataset-main.zip recorded at Lake Constance. Our experiments show that the pitch and roll estimator provides accurate results in comparison to an Xsens IMU sensor. We can further improve the pitch and roll estimation by sensor fusion with a gyroscope. The algorithm is available in its implementation as a ROS node33https://github.com/dionysos4/water_surface_detector.

Grounding language to the visual observations of a navigating agent can be performed using off-the-shelf visual-language models pretrained on Internet-scale data (e.g., image captions). While this is useful for matching images to natural language descriptions of object goals, it remains disjoint from the process of mapping the environment, so that it lacks the spatial precision of classic geometric maps. To address this problem, we propose VLMaps, a spatial map representation that directly fuses pretrained visual-language features with a 3D reconstruction of the physical world. VLMaps can be autonomously built from video feed on robots using standard exploration approaches and enables natural language indexing of the map without additional labeled data. Specifically, when combined with large language models (LLMs), VLMaps can be used to (i) translate natural language commands into a sequence of open-vocabulary navigation goals (which, beyond prior work, can be spatial by construction, e.g., “in between the sofa and the TV” or “three meters to the right of the chair”) directly localized in the map, and (ii) can be shared among multiple robots with different embodiments to generate new obstacle maps on-the-fly (by using a list of obstacle categories). Extensive experiments carried out in simulated and real-world environments show that VLMaps enable navigation according to more complex language instructions than existing methods. Videos are available at https://vlmaps.github.io.

We consider how to most efficiently leverage teleoperator time to collect data for learning robust image-based value functions and policies for sparse reward robotic tasks. To accomplish this goal, we modify the process of data collection to include more than just successful demonstrations of the desired task. Instead we develop a novel protocol that we call Visual Backtracking Teleoperation (VBT), which deliberately collects a dataset of visually similar failures, recoveries, and successes. VBT data collection is particularly useful for efficiently learning accurate value functions from small datasets of image-based observations. We demonstrate VBT on a real robot to perform continuous control from image observations for the deformable manipulation task of T-shirt grasping. We find that by adjusting the data collection process we improve the quality of both the learned value functions and policies over a variety of baseline methods for data collection. Specifically, we find that offline reinforcement learning on VBT data outperforms standard behavior cloning on successful demonstration data by 13 % when both methods are given equal-sized datasets of 60 minutes of data from the real robot.

Motivated by the intuitive understanding humans have about the space of possible interactions, and the ease with which they can generalize this understanding to previously unseen scenes, we develop an approach for learning ‘visual affordances’. Given an input image of a scene, we infer a distribution over plausible future states that can be achieved via interactions with it. To allow predicting diverse plausible futures, we discretize the space of continuous images with a VQ-VAE and use a Transformer-based model to learn a conditional distribution in the latent embedding space. We show that these models can be trained using large-scale and diverse passive data, and that the learned models exhibit compositional generalization to diverse objects beyond the training distribution. We evaluate the quality and diversity of the generations, and demonstrate how the trained affordance model can be used for guiding exploration during visual goal-conditioned policy learning in robotic manipulation.

We study six-dimensional (6D) perceptive peg-in-hole problem for practical connector insertion task in this paper. To enable the manipulator system to handle different types of pegs in complex environment, we develop a perceptive robotic assembly system that utilizes an in-hand RGB-D camera for peg-in-hole with multiple types of pegs. The proposed framework addresses the critical hole detection and pose estimation problem through combining the learning-based detection with model-based pose estimation strategies. By exploiting the structure of the peg-in-hole task, we consider a rectangle-shape based characterization for modeling the candidate socket. Such a characterization allows us to design simple learning-based methods to detect and estimate the 6D pose of the target socket that balances between processing speed and accuracy. To validate our method, we test the performance of the proposed perceptive peg-in-hole solution using a KUKA iiwa7 robotic arm to accomplish the socket insertion task with two types of practical sockets (RJ45/HDMI). Without the need of additional search, our method achieves an acceptable success rate in the connector insertion tasks. The results confirm the reliability of our method and show that our method is suitable for real world application.

Navigating dynamic environments requires the robot to generate collision-free trajectories and actively avoid moving obstacles. Most previous works designed path planning algorithms based on one single map representation, such as the geometric, occupancy, or ESDF map. Although they have shown success in static environments, due to the limitation of map representation, those methods cannot reliably handle static and dynamic obstacles simultaneously. To address the problem, this paper proposes a gradient-based B-spline trajectory optimization algorithm utilizing the robot's onboard vision. The depth vision enables the robot to track and represent dynamic objects geometrically based on the voxel map. The proposed optimization first adopts the circle-based guide-point algorithm to approximate the costs and gradients for avoiding static obstacles. Then, with the vision-detected moving objects, our receding-horizon distance field is simultaneously used to prevent dynamic collisions. Finally, the iterative re-guide strategy is applied to generate the collision-free trajectory. The simulation and physical experiments prove that our method can run in real-time to navigate dynamic environments safely.

In this paper, we examine the problem of visibility-aware robot navigation among movable obstacles (VANAMO). A variant of the well-known NAMO robotic planning problem, VANAMO puts additional visibility constraints on robot motion and object movability. This new problem formulation lifts the restrictive assumption that the map is fully visible and the object positions are fully known. We provide a formal definition of the VANAMO problem and propose the Look and Manipulate Backchaining (LAMB) algorithm for solving such problems. Lamb has a simple vision-based interface that makes it more easily transferable to real-world robot applications and scales to the large 3D environments. To evaluate Lamb, we construct a set of tasks that illustrate the complex interplay between visibility and object movability that can arise in mobile base manipulation problems in unknown environments. We show that Lamb outperforms NAMO and visibility-aware motion planning approaches as well as simple combinations of them on complex manipulation problems with partial observability.

To assist robots in teleoperation tasks, haptic rendering which allows human operators access a virtual touch feeling has been developed in recent years. Most previous haptic rendering methods strongly rely on data collected by tactile sensors. However, tactile data is not widely available for robots due to their limited reachable space and the restrictions of tactile sensors. To eliminate the need for tactile data, in this paper we propose a novel method named as Vis2Hap to generate haptic rendering from visual inputs that can be obtained from a distance without physical interaction. We take the surface texture of objects as key cues to be conveyed to the human operator. To this end, a generative model is designed to simulate the roughness and slipperiness of the object's surface. To embed haptic cues in Vis2Hap, we use height maps from tactile sensors and spectrograms from friction coefficients as the intermediate outputs of the generative model. Once Vis2Hap is trained, it can be used to generate height maps and spectrograms of new surface textures, from which a friction image can be obtained and displayed on a haptic display. The user study demonstrates that our proposed Vis2Hap method enables users to access a realistic haptic feeling similar to that of physical objects. The proposed vision-based haptic rendering has the potential to enhance human operators' perception of the remote environment and facilitate robotic manipulation.

Every autonomous driving dataset has a different configuration of sensors, originating from distinct geographic regions and covering various scenarios. As a result, 3D detectors tend to overfit the datasets they are trained on. This causes a drastic decrease in accuracy when the detectors are trained on one dataset and tested on another. We observe that lidar scan pattern differences form a large component of this reduction in performance. We address this in our approach, SEE-VCN, by designing a novel viewer-centred surface completion network (VCN) to complete the surfaces of objects of interest within an unsupervised domain adaptation framework, SEE [1]. With SEE-VCN, we obtain a unified representation of objects across datasets, allowing the network to focus on learning geometry, rather than overfitting on scan patterns. By adopting a domain-invariant representation, SEE-VCN can be classed as a multi-target domain adaptation approach where no annotations or re-training is required to obtain 3D detections for new scan patterns. Through extensive experiments, we show that our approach outperforms previous domain adaptation methods in multiple domain adaptation settings. Our code and data are available at https://github.com/darrenjkt/SEE-VCN.

The waterdrops on windshields during driving can cause severe visual obstructions, which may lead to car accidents. Meanwhile, the waterdrops can also degrade the performance of a computer vision system in autonomous driving. To address these issues, we propose an attention-based framework that fuses the spatio-temporal representations from multiple frames to restore visual information occluded by waterdrops. Due to the lack of training data for video waterdrop removal, we propose a large-scale synthetic dataset with simulated waterdrops in complex driving scenes on rainy days. To improve the generality of our proposed method, we adopt a cross-modality training strategy that combines synthetic videos and real-world images. Extensive experiments show that our proposed method can generalize well and achieve the best waterdrop removal performance in complex real-world driving scenes.

Recently, a growing number of work design unsupervised paradigms for point cloud processing to alleviate the limitation of expensive manual annotation and poor transferability of supervised methods. Among them, CrossPoint follows the contrastive learning framework and exploits image and point cloud data for unsupervised point cloud understanding. Although the promising performance is presented, the unbalanced architecture makes it unnecessarily complex and inefficient. For example, the image branch in CrossPoint is ~8.3x heavier than the point cloud branch leading to higher complexity and latency. To address this problem, in this paper, we propose a lightweight Vision-and-Pointcloud Transformer (ViPFormer) to unify image and point cloud processing in a single architecture. ViPFormer learns in an unsupervised manner by optimizing intra-modal and cross-modal contrastive objectives. Then the pretrained model is transferred to various downstream tasks, including 3D shape classification and semantic segmentation. Experiments on different datasets show ViPFormer surpasses previous state-of-the-art unsupervised methods with higher accuracy, lower model complexity and runtime latency. Finally, the effectiveness of each component in ViPFormer is validated by extensive ablation studies. The implementation of the proposed method is available at https://github.com/auniquesun/ViPFormer.

We present Visual Navigation and Locomotion over obstacles (ViNL), which enables a quadrupedal robot to navigate unseen apartments while stepping over small obstacles that lie in its path (e.g., shoes, toys, cables), similar to how humans and pets lift their feet over objects as they walk. ViNL consists of: (1) a visual navigation policy that outputs linear and angular velocity commands that guides the robot to a goal coordinate in unfamiliar indoor environments; and (2) a visual locomotion policy that controls the robot's joints to avoid step-ping on obstacles while following provided velocity commands. Both the policies are entirely ‘model-free’, i.e. sensors-to-actions neural networks trained end-to-end. The two are trained independently in two entirely different simulators and then seamlessly co-deployed by feeding the velocity commands from the navigator to the locomotor, entirely ‘zero-shot’ (without any co-training). While prior works have developed learning methods for visual navigation or visual locomotion, to the best of our knowledge, this is the first fully learned approach that leverages vision to accomplish both (1) intelligent navigation in new environments, and (2) intelligent visual locomotion that aims to traverse cluttered environments without disrupting obstacles. On the task of navigation to distant goals in unknown environments, ViNL using just egocentric vision significantly outperforms prior work on robust locomotion using privileged terrain maps (+32.8% success and -4.42 collisions per meter). Additionally, we ablate our locomotion policy to show that each aspect of our approach helps reduce obstacle collisions. Videos and code at http://www.joannetruong.com/projects/vinl.html.

Learning diverse skills is one of the main challenges in robotics. To this end, imitation learning approaches have achieved impressive results. These methods require explicitly labeled datasets or assume consistent skill execution to enable learning and active control of individual behaviors, which limits their applicability. In this work, we propose a cooperative adversarial method for obtaining single versatile policies with controllable skill sets from unlabeled datasets containing diverse state transition patterns by maximizing their discriminability. Moreover, we show that by utilizing unsupervised skill discovery in the generative adversarial imitation learning framework, novel and useful skills emerge with successful task fulfillment. Finally, the obtained versatile policies are tested on an agile quadruped robot called Solo 8 and present faithful replications of diverse skills encoded in the demonstrations.

Specialized motions such as jumping are often achieved on quadruped robots by solving a trajectory optimization problem once and executing the trajectory using a tracking controller. This approach is in parallel with Model Predictive Control (MPC) strategies that commonly control regular gaits via online re-planning. In this work, we present a nonlinear MPC (NMPC) technique that unlocks on-the-fly replanning of specialized motion skills and regular locomotion within a unified framework. The NMPC reasons about a hybrid kinodynamic model, and is solved using a variant of a constrained Differential Dynamic Programming (DDP) solver. The proposed NMPC enables the robot to perform a variety of agile skills like jumping, bounding, and trotting, and the rapid transition between them. We evaluated the proposed algorithm with three challenging motion sequences that combine multiple agile skills, on two quadruped platforms, Unitree A1, and MIT Mini Cheetah, showing its effectiveness and generality.

Here we describe how a 10-gram Flapping-Wing Micro Aerial Vehicle (FWMAV) was able to perform an automatic trajectory tracking task based on a vector field method. In this study, the desired heading was provided by a vector field which was computed depending on the desired trajectory. The FWMAV's heading was changed by a rear steering mechanism. This rear mechanism simultaneously (i) tenses one wing and relaxes the opposite wing, and (ii) moves the rudder in the same direction as the wing is relaxed. Due to the complex dynamics, system identification methods were used to identify simple linear models using a set of dedicated free flight tests. This yaw and roll simple models help to adjust the yaw controller and the inner loop roll controller. The experimental results obtained here show that a time-independent vector field-based strategy is robust to various initial position and/or speed conditions. The task of tracking circular and 8-shaped trajectories was accomplished successfully over tens of meters.

A compliant control model based on reinforcement learning (RL) is proposed to allow robots to interact with the environment more effectively and autonomously execute force control tasks. The admittance model learns an optimal adjustment policy for interactions with the external environment using RL algorithms. The model combines energy consumption and trajectory tracking of the agent state using a cost function. Therein, an Unmanned Aerial Vehicle (UAV) can operate stably in unknown environments where interaction forces exist. Furthermore, the model ensures that the interaction process is safe, comfortable, and flexible while protecting the external structures of the UAV from damage. To evaluate the model performance, we verified the approach in a simulation environment using a UAV in three external force scenes. We also tested the model across different UAV platforms and various low-level control parameters, and the proposed approach provided the best results.