ICRA 2023 论文列表

Visual place recognition (VPR) is one of the research hotspots in robotics, which uses visual information to locate robots. Recently, the hierarchical two-stage VPR methods have become popular in this field due to the trade-off between accuracy and efficiency. These methods retrieve the top-k candidate images using the global features in the first stage, then re-rank the candidates by matching the local features in the second stage. However, they usually require additional al-gorithms (e.g. RANSAC) for geometric consistency verification in re-ranking, which is time-consuming. Here we propose a Dynamically Aligning Local Features (DALF) algorithm to align the local features under spatial constraints. It is significantly more efficient than the methods that need geometric consistency verification. We present a unified network capable of extracting global features for retrieving candidates via an aggregation module and aligning local features for re-ranking via the DALF alignment module. We call this network AANet. Meanwhile, many works use the simplest positive samples in triplet for weakly supervised training, which limits the ability of the network to recognize harder positive pairs. To address this issue, we propose a Semi-hard Positive Sample Mining (ShPSM) strategy to select appropriate hard positive images for training more robust VPR networks. Extensive experiments on four benchmark VPR datasets show that the proposed AANet can outperform several state-of-the-art methods with less time consumption. The code is released at https://github.com/Lu-Feng/AANet.

The real-time dynamic environment perception has become vital for autonomous robots in crowded spaces. Although the popular voxel-based mapping methods can efficiently represent 3D obstacles with arbitrarily complex shapes, they can hardly distinguish between static and dynamic obstacles, leading to the limited performance of obstacle avoidance. While plenty of sophisticated learning-based dynamic obstacle detection algorithms exist in autonomous driving, the quad-copter's limited computation resources cannot achieve real-time performance using those approaches. To address these issues, we propose a real-time dynamic obstacle tracking and mapping system for quadcopter obstacle avoidance using an RGB-D camera. The proposed system first utilizes a depth image with an occupancy voxel map to generate potential dynamic obstacle regions as proposals. With the obstacle region proposals, the Kalman filter and our continuity filter are applied to track each dynamic obstacle. Finally, the environment-aware trajectory prediction method is proposed based on the Markov chain using the states of tracked dynamic obstacles. We implemented the proposed system with our custom quadcopter and navigation planner. The simulation and physical experiments show that our methods can successfully track and represent obstacles in dynamic environments in real-time and safely avoid obstacles.

Borrowing elementary ideas from solid mechanics and differential geometry, this presentation shows that the volume swept by a regular solid undergoing a wide class of volume-preserving deformations induces a rather natural metric structure with well-defined and computable geodesics on its configuration space. This general result applies to concrete classes of articulated objects such as robot manipulators, and we demonstrate as a proof of concept the computation of geodesic paths for a free flying rod and planar robotic arms as well as their use in path planning with many obstacles.

Powered lower-limb exoskeletons have been shown to assist and augment walking, but most such devices do not currently have the ability to explicitly accommodate a stumble perturbation. A major challenge in doing so is identifying a stumble event and selecting in real-time which recovery strategy (elevating or lowering) to employ, particularly since the exoskeleton should ideally select the same strategy selected by the user. In order to do so, the authors conducted experiments involving five young, healthy adults wearing a knee exoskeleton. Each participant underwent a stumble experiment in order to collect an exoskeleton sensor dataset of stumbles throughout swing phase, which was used for stumble detection and recovery strategy identification algorithm development and testing. Overall, the proposed detection and identification algorithms provide improved accuracy with fewer required sensors relative to previous works, and were tested on the largest exoskeleton sensor stumble dataset to date, showing the feasibility of such algorithms for real-time implementation, which is an essential first step in developing lower-limb assistive devices that are robust to stumbles.

Flight is an energetically expensive task. While aerial insects can effortlessly fly through natural environments, achieving power autonomous flights in insect-scale robots remains a major challenge. In prior works, we developed soft-actuated insect-scale aerial robots that demonstrated unique capabilities such as in-flight collision recovery and somersaults. However, the soft dielectric elastomer actuators (DEAs) have low efficiency (< 20%) and require a high driving voltage (>600 V). These properties represent formidable obstacles for soft aerial robots to achieve power autonomous flights. In this work, we developed a 127 mg boost circuit that can convert a 7.7 V DC input into a 600 V and 400 Hz output for driving a 120 mg DEA. It has an equivalent capacitance and resistance of 20 nF and 5 \mathbf{k}\Omega\mathbf{k}\Omega, respectively. The DEA is assembled into a 158 mg aerial robot, which can demonstrate liftoff while carrying the boost circuit as a payload. Although the robot remains tethered to an off-board power supply, this result represents a first step towards achieving power autonomy in soft aerial robots.

As a permanent solution for patients who cannot contract their urinary bladder, an artificial detrusor muscle appears a higher outcome approach compared to current sacral neurostimulators featured by severe long-term side effects. In this paper, a novel soft robotic detrusor is presented to overcome the limitations of the state-of-the-art solutions. It is based on two identical origami-based hydraulic actuators, which completely surround the bladder and contract upon water aspiration. Design, manufacturing, and experimental characterization both in terms of contraction capabilities and voiding efficiency on ex vivo swine bladders are reported for two different origami geometries, as well as a proof-of-concept implementation of an autonomous driving circuit as control unit. Results from assisted urination tests outlined very good performances proving an active voiding efficiency of the hydraulic soft robotic detrusor equal to 84.8% \pm 7.4\%\pm 7.4\% in simulated environment.

This paper presents a generic power wheelchair dynamic model. As a first contribution, this paper proposes to use a generic model composed of a geometric model and a lumped model in order to be compliant with a wide range of existing commercially available wheelchairs. In this model, a set of essential parameters are enough to accurately replicate the dynamic behavior of a wheelchair. As a second contribution, this paper presents an identification method of a n-wheel type power wheelchair. The presented model is restricted to the sagittal plane only, which is sufficient to study the reliability of the identification and validation methods. Moreover, a Motion Cueing Algorithm based on the proposed model controls a simulator mechanical platform. The generic model has been then validated through a user study with 18 able-bodied participants evaluating the self-motion perception with our multisensory power wheelchair driving simulator. Results show that the simplified model is sufficient to provide accurate sensations to the user with respect to their experience while driving a power wheelchair.

Predicting 3D human poses in real-world scenarios, also known as human pose forecasting, is inevitably subject to noisy inputs arising from inaccurate 3D pose estimations and occlusions. To address these challenges, we propose a diffusion-based approach that can predict given noisy observations. We frame the prediction task as a denoising problem, where both observation and prediction are considered as a single sequence containing missing elements (whether in the observation or prediction horizon). All missing elements are treated as noise and denoised with our conditional diffusion model. To better handle long-term forecasting horizon, we present a temporal cascaded diffusion model. We demonstrate the benefits of our approach on four publicly available datasets (Human3.6M, HumanEva-I, AMASS, and 3DPW), outperforming the state-of-the-art. Additionally, we show that our framework is generic enough to improve any 3D pose prediction model as a pre-processing step to repair their inputs and a post-processing step to refine their outputs. The code is available online: https://github.com/vita-epfl/DePOSit.

State minimization of combinatorial filters is a fundamental problem that arises, for example, in building cheap, resource-efficient robots. But exact minimization is known to be NP-hard. This paper conducts a more nuanced analysis of this hardness than up till now, and uncovers two factors which contribute to this complexity. We show each factor is a distinct source of the problem's hardness and are able, thereby, to shed some light on the role played by (1) structure of the graph that encodes compatibility relationships, and (2) determinism-enforcing constraints. Just as a line of prior work has sought to introduce additional assumptions and identify sub-classes that lead to practical state reduction, we next use this new, sharper understanding to explore special cases for which exact minimization is efficient. We introduce a new algorithm for constraint repair that applies to a large sub-class of filters, subsuming three distinct special cases for which the possibility of optimal minimization in polynomial time was known earlier. While the efficiency in each of these three cases previously appeared to stem from seemingly dissimilar properties, when seen through the lens of the present work, their commonality now becomes clear. We also provide entirely new families of filters that are efficiently reducible.

One of the biggest problems with soft robots is precisely the fact that they are soft. Indeed the softer they are, the less force they can exert on the environment. Researchers have proposed a number of stiffening methods, but all of them have drawbacks, such as locking the shape of the device in a way that precludes further adjustments. In this paper we propose a stiffening method inspired by the internal structure of the mammalian penis. The soft actuation chamber is divided into small compartments that trap the actuation fluid, leading to locally amplified pressure increase under certain conditions. At the same time, the proposed solution does not affect the actuation mechanism, allowing the actuator to be adjusted in one direction just as if it was in non-stiffened mode, while offering a stiff response in the opposite direction. Our prototype achieves an increase in stiffening of approximately a factor of two. The paper describes the concept, the mathematical justification of the working principle, the prototype design, its implementation and our experimental results.

We consider the problem of planning the motion of a drone equipped with a robotic arm, tasked with bringing its end-effector up to many (150+) targets in a fruit tree; to inspect every piece of fruit, for example. The task is complicated by the intersection of a version of Neighborhood TSP (to find an optimal order and a pose to visit every target), and a robotic motion-planning problem through a planning space that features numerous cavities and narrow passages that confuse common techniques. In this contribution, we present a framework that decomposes the problem into two stages: planning approach paths for every target, and quickly planning between the start points of those approach paths. Then, we compare our approach by simulation to a more straightforward method based on multiquery planning, showing that our approach outperforms it in both time and solution cost.

A congestion-aware path planning method is pre-sented for mobile robots during long-term deployment in human occupied environments. With known spatial-temporal crowd patterns, the robot will navigate to its destination via less congested areas. Traditional traffic-aware routing methods do not consider spatial-temporal anomalies of macroscopic crowd behaviour that can deviate from the predicted crowd spatial distribution. The proposed method improves long-term path planning adaptivity by integrating a partially updated memory (PUM) model that utilizes observed anomalies to generate a multi-layer crowd density map to improve estimation accuracy. Using this map, we are able to generate a path that has less chance to encounter the crowded areas. Simulation results show that our method outperforms the benchmark congestion-aware routing method in terms of reducing the probability of robot's proximity to dense crowds.

Future conceptions of agile, just-in-time fabrication, lean and “smart” manufacturing, and a host of allied processes that exploit advanced automation, depend in part on realizing improvements in logistics planning. The present paper hypothesizes that the key to improving flexibility will be the inclusion of sophisticated, time-correlated stochastic models of demand—whether that be demand by end-user consumers directly, or by other down-stream processes. Such dynamic models of demand, unfortunately, can greatly increase the space in which planning occurs when treated, as is common for planning under uncertainty, via the Markov Decision Processes formulation. To tackle this challenge, we identify three aspects that we postulate appear as commonalities in many logistics settings. They lead to an approach for approximate reduction of the planning problem via causal decoupling, which gives a spectrum of solutions where weakening time correlations affords faster optimization. Empirical results on small case studies —in lean manufacturing and commodity routing—show that retaining some limited (but non-zero) amount of temporal structure can provide a useful compromise between quality of the solution obtained and computation required.

Teleoperation of robots in space is challenging due to high latency and limited workspace visibility. Previously, the Interactive Planning and Supervised Execution (IPSE) and Augmented Virtuality systems were developed to reduce failure risk. These tools were visualized on a 3D da Vinci surgical console and operated using the da Vinci manipulators or visualized on conventional monitors and operated with a keyboard and mouse. Experimental studies indicated operator preference for the latter. In this work, we develop a 3D virtual reality (VR) interface for IPSE, implemented on a Meta Quest 2 head-mounted display (HMD), and evaluate it against the prior 2D, keyboard-and-mouse-based interface. The results demonstrate improved operator load with the 3D VR interface, with no decrease in task performance, while also providing cost and portability benefits compared to the conventional 2D interface.

Teaching complex manipulation skills, such as folding garments, to a bi-manual robot is a very challenging task, which is often tackled through learning from demon-stration. The few datasets of garment-folding demonstrations available nowadays to the robotics research community have been either gathered from human demonstrations or generated through simulation. The former have the great difficulty of perceiving both cloth state and human action as well as transferring them to the dynamic control of the robot, while the latter require coding human motion into the simulator in open loop, i.e., without incorporating the visual feedback naturally used by people, resulting in far-from-realistic movements. In this article, we present an accurate dataset of human cloth folding demonstrations. The dataset is collected through our novel virtual reality (VR) framework, based on Unity's 3D platform and the use of an HTC Vive Pro system. The framework is capable of simulating realistic garments while allowing users to interact with them in real time through handheld controllers. By doing so, and thanks to the immersive experience, our framework permits exploiting human visual feedback in the demonstrations while at the same time getting rid of the difficulties of capturing the state of cloth, thus simplifying data acquisition and resulting in more realistic demonstrations. We create and make public a dataset of cloth manipulation sequences, whose cloth states are semantically labeled in an automatic way by using a novel low-dimensional cloth representation that yields a very good separation between different cloth configurations.

Pairwise point cloud registration is a critical task for many applications, which heavily depends on finding correct correspondences from the two point clouds. However, the low overlap between input point clouds causes the registration to fail easily, leading to mistaken overlapping and mismatched correspondences, especially in scenes where non-overlapping regions contain similar structures. In this paper, we present a unified bird's-eye view (BEV) model for jointly learning of 3D local features and overlap estimation to fulfill pairwise registration and loop closure. Feature description is performed by a sparse UNet-like network based on BEV representation, and 3D keypoints are extracted by a detection head for 2D locations, and a regression head for heights. For overlap detection, a cross-attention module is applied for interacting contextual information of input point clouds, followed by a classification head to estimate the overlapping region. We evaluate our unified model extensively on the KITTI dataset and Apollo-SouthBay dataset. The experiments demonstrate that our method significantly outperforms existing methods on overlap estimation, especially in scenes with small overlaps. It also achieves top registration performance on both datasets in terms of translation and rotation errors.

Motion planning for mobile robot platforms is one of the long-established research fields in robotics. In this paper, we propose a trajectory planner for mobile holonomic robots to steer non-holonomic conventional passive wheelchairs in dynamic environments. The challenges to overcome when steering a wheelchair are to find smooth feasible trajectories, maintain a fast reactive response to dynamic obstacles and to satisfy a set of additional constraints such as limiting physical forces acting on the wheelchair occupants. Our approach is a variant of the timed-elastic-bands (TEB) planner, which includes a footprint of the wheelchair during optimization, and generates a steering angle which is then consumed by an arm controller to actuate the relative orientation between the wheelchair and the mobile platform. This is realized by posing new non-holonomic and kinodynamic constraints on the TEB planner and an implementation of a suitable real-time dual-arm controller for executing steering commands. We demonstrate our results based on a TEB baseline comparison in simulation using functional models of our robot HoLLiE and a wheelchair.

The requirements of modern production systems together with more advanced robotic technologies have fostered the integration of teams comprising humans and autonomous robots. While this integration has the potential to provide various benefits, it also raises questions about how to effectively manage these teams, taking into account the different characteristics of the agents involved. This paper presents a framework for task allocation in a human multi-robot collaborative scenario. The proposed solution combines an optimal offline allocation with an online reallocation strategy which accounts for inaccuracies of the offline plan and/or unforeseen events, human subjective preferences and cost of task switching. Experiments with two manipulators cooperating with a human operator in a box filling task are presented.

This work presents a novel hybrid rigid-soft continuum manipulator, which integrates high-resolution tactile sensing in a form factor that is forceful, compliant, inherently safe, and easily controllable. We utilize a hybrid approach motivated by scale-invariant principles to fuse the rigid and soft design domains while addressing their respective challenges. We use Euler-Bernoulli beam theory and geometric inference to design and develop a novel variant of folded flexure hinge (FFH) compliant mechanism, the variable area moment of inertia folded flexure hinge (VAFFH), which deforms logarithmically along its length and thus yields first-order scale-invariant grasp behavior. Finally, we characterize the forcefulness of the manipulator and demonstrate its compliance, adaptability, and tactile sensing capabilities in selected tasks.