IROS 2022 论文列表

Undulation is the most common gait generated by legless creatures, which enables their robust and efficient locomotion in various environments. Such advantages inspired the control design of many kinds of locomotion robots. Despite their technical details, most of them realize the undulation gait via tracking predetermined trajectories called serpenoid curves, which are a group of sinusoidal waveforms with specified phase differences. This technique, however, sounds quite redundant in terms of sensing and control. Here, we investigate the research question: whether the sinusoidal waveform is necessary to be encoded in the control signal to make the whole body an “S-shape”? We use a 4-link rigid body dynamics model as a simple example, by which numerical simulations are conducted. Together with theoretical analysis, we show that undulation gait emerges naturally based on embodied position controller and filter, where binary actuation torques are required only. Our results not only discover locomotion mechanisms for significantly reducing the sensing and control requirement of generating artificial undulation gait, but also provide additional understandings for biological systems from the mechanical engineering point of view.

Embedding an optimization process has been explored for imposing efficient and flexible policy structures. Existing work often build upon nonlinear optimization with explicitly iteration steps, making policy inference prohibitively expensive for online learning and real-time control. Our approach embeds a linear-quadratic-regulator (LQR) formulation with a Koopman representation, thus exhibiting the tractability from a closed-form solution and richness from a non-convex neural network. We use a few auxiliary objectives and reparameterization to enforce optimality conditions of the policy that can be easily integrated to standard gradient-based learning. Our approach is shown to be effective for learning policies rendering an optimality structure and efficient reinforcement learning, including simulated pendulum control, 2D and 3D walking, and manipulation for both rigid and deformable objects. We also demonstrate real world application in a robot pivoting task.

Tendon-driven flexible endoscopic surgical robots have been developed to access narrow curved paths without incision. Robot shape information is essential for precise control and to prevent unwanted tissue damage. In this paper, we propose a joint angle sensing method using coiled soft sensors to estimate the shape of the hyperredundant manipulator, which is commonly used in flexible endoscopic surgical robots. The soft sensors can be fabricated with small size and are highly stretchable, such that by being pre-stretched, they can be integrated between individual joints, maintain a center hollow, and sense both compression and extension. The pre-stretch length is experimentally selected by using the sensor linearity to maximize the potential sensitivity. We validated the proposed design using a two-degree of freedom (DOF) single joint manipulator by implementing two sensors; sensors at all joints could sense joint angle independently and simultaneously with a root-mean-square error (RMSE) less than 2.53°. Based on the proposed method, a two-DOF configuration of the hyperredundant manipulator that can be used in real applications was achieved, following a constant curvature model in real time with values RMSE of 2.30° and 2.63°, for pitch and yaw joint angle respectively.

This paper introduces a new method for robot motion planning and navigation in uneven environments through a surfel representation of underlying point clouds. The proposed method addresses the shortcomings of state-of-the-art navigation methods by incorporating both kinematic and physical constraints of a robot with standard motion planning algorithms (e.g., those from the Open Motion Planning Library), thus enabling efficient sampling-based planners for challenging uneven terrain navigation on raw point cloud maps. Unlike techniques based on Digital Elevation Maps (DEMs), our novel surfel-based state-space formulation and implementation are based on raw point cloud maps, allowing for the modeling of overlapping surfaces such as bridges, piers, and tunnels. Experimental results demonstrate the robustness of the proposed method for robot navigation in real and simulated unstructured environments. The proposed approach also optimizes planners' performances by boosting their success rates up to 5x for challenging unstructured terrain planning and navigation, thanks to our surfel-based approach's robot constraint-aware sampling strategy. Finally, we provide an open-source implementation of the proposed method to benefit the robotics community.

Perceiving the surrounding environment is crucial for autonomous mobile robots. An elevation map provides a memory-efficient and simple yet powerful geometric represen-tation of the terrain for ground robots. The robots can use this information for navigation in an unknown environment or perceptive locomotion control over rough terrain. Depending on the application, various post processing steps may be incorpo-rated, such as smoothing, inpainting or plane segmentation. In this work, we present an elevation mapping pipeline leveraging GPU for fast and efficient processing with additional features both for navigation and locomotion. We demonstrated our map-ping framework through extensive hardware experiments. Our mapping software was successfully deployed for underground exploration during DARPA Subterranean Challenge and for various experiments of quadrupedal locomotion.

Soft robotic actuators are safe and adaptable devices with inherent compliance, which makes them attractive for manipulating delicate and complex objects. Researchers have integrated stiff materials into soft actuators to increase their force capacity and direct their deformation. However, these embedded materials have largely been pre-prescribed and static, which constrains the actuators to a predetermined range of motion. In this work, electroadhesive (EA) clutches integrated on a single-chamber soft pneumatic actuator (SPA) provide local programmable stiffness modulation to control the actuator deformation. We show that activating different clutch patterns inflates a silicone membrane into pyramidal, round, and plateau shapes. Curvatures from these shapes are combined during actuation to apply forces on both a 3.7 g and 820 g object along five different degrees of freedom (DoF). The actuator workspace is up to 12 mm for light objects. Clutch deactivation, which results in local elastomeric expansion, rapidly applies forces up to 3.2 N to an object resting on the surface and launches a 3.7 g object in controlled directions. The actuator also rotates a heavier, 820 g, object by 5 degrees and rapidly restores it to horizontal alignment after clutch deactivation. This actuator is fully powered by a 5 V battery, AA battery, DC-DC transformer, and 4.5 V (63 g) DC air pump. These results demonstrate a first step towards realizing a soft actuator with high DoF shape change that preserves the inherent benefits of pneumatic actuation while gaining the electrical controllability and strength of EA clutches. We envision such a system supplying human contact forces in the form of a low-profile sit-to-stand assistance device, bed-ridden patient manipulator, or other ergonomic mechanism.

Electro-adhesive clutches have become effective tools for variable stiffness functions in many robotic systems due to their light weight, high speed and strong brake force. In this paper, we present a novel, tubular design of an electro-adhesive clutch. Our clutch consists of flexible electrode sheets rolled into a tubular structure. This design allows encapsulating large electrode areas in a compact size for strong brake force. Additionally, the tubular structure acts as a guide for directional sliding without external guides. The structure also ensures that the electrode surfaces are encapsulated, preventing the accumulation of dust and thus leading to reliable performance. This structure is therefore an improvement over the commonly used planar designs. The characterization of the electro-adhesive tubular clutch shows that the frictional force increases with the increase of the electrode contact area, the decrease of the roll diameter and the dielectric layer thickness. A retractable tubular clutch is made by fixing an elastic cable along the clutch axis and achieves a stiffness change factor up to 260. Applications of this retractable clutch in robotics to achieve variable stiffness are demonstrated in two systems: a tensegrity structure and a wing skeleton. Changes in stiffness by 13.2 and 30.2 times are achieved for the two systems, respectively. The proposed tubular clutch is an effective means of achieving variable stiffness, particularly in the case of robotic systems that transmit forces through tensioned cables.

Egocentric vision is an emerging topic, which has demonstrated great potential in assistive healthcare scenarios, ranging from human-centric behavior analysis to personal social assistance. Within this field, due to the heterogeneity of visual perception from first-person views, egocentric pose estimation is one of the most significant prerequisites for enabling various downstream applications. However, existing methods for egocentric pose estimation mainly focus on predicting the pose represented in the camera coordinates from a single image, which ignores the latent cues in the temporal domain and results in less accuracy. In this paper, we propose Ego+X, an egocentric vision based system for 3D canonical pose estimation and human-centric social interaction characterization. Our system is composed of two head-mounted egocentric cameras, where one is faced downwards and the other looks outwards. By leveraging the global context provided by visual SLAM, we first propose Ego-Glo for spatial-accurate and temporal-consistent egocentric 3D pose estimation in the canonical coordinate system. With the help of an egocentric camera looking outwards, we then propose Ego-Soc by extending Ego-Glo to various social interaction tasks, e.g., object detection and human-human interaction. Quantitative and qualitative experiments have been conducted to demonstrate the effectiveness of our proposed Ego+X.

A key challenge in learning to perform manipulation tasks is selecting a suitable skill representation. While specific skill representations are often easier to learn, they are often only suitable for a narrow set of tasks. In most prior works, roboticists manually provide the robot with a suitable skill representation to use e.g. a neural network or DMPs. By contrast, we propose to allow the robot to select the most appropriate skill representation for the underlying task. Given the large space of skill representations, we utilize a single demonstration to select a small set of potential task-relevant representations. This set is then further refined using reinforcement learning to select the most suitable skill representation. Experiments in both simulation and real world show how our proposed approach leads to improved sample efficiency and enables directly learning on the real robot.

When generating robot motions based on instructions such as cooking recipes, ambiguity of the instructions and lack of necessary information are problematic for the robot. To solve this problem, we propose an efficient motion planning approach for a dual-arm robot by constructing a graph repre-senting a motion sequence based on a recipe consisting of verbal instructions and cooking images. A functional unit is generated based on the linguistic instructions in the recipe. Since most recipes lack the necessary information for executing the motion, we first consider extracting the information about the cooking motion like cutting from the food images of the recipe and supplementing it. In addition, to supplement the actions that humans perform unconsciously, we generate functional units for actions not explicitly mentioned in the recipe based on the current situation of the cooking process, and then connect them to the functional units generated from the recipe. Moreover, during the connection we consider the motion of the robot's arms in parallel for an efficient execution of the recipe, similar to those of a human. Through experiments, we demonstrate that for a given recipe, the proposed method can be used to generate a cooking sequence with the supplementary information needed, and executed by a dual-arm robot. The results show that the proposed method is effective and can simplify robot teaching in cooking tasks.

Accurate moving object segmentation is an es-sential task for autonomous driving. It can provide effective information for many downstream tasks, such as collision avoidance, path planning, and static map construction. How to effectively exploit the spatial-temporal information is a critical question for 3D LiDAR moving object segmentation (LiDAR-MOS). In this work, we propose a novel deep neural network exploiting both spatial-temporal information and different representation modalities of LiDAR scans to improve LiDAR-MOS performance. Specifically, we first use a range image-based dual-branch structure to separately deal with spatial and temporal information that can be obtained from sequential LiDAR scans, and later combine them using motion-guided attention modules. We also use a point refinement module via 3D sparse convolution to fuse the information from both LiDAR range image and point cloud representations and reduce the artifacts on the borders of the objects. We verify the effectiveness of our proposed approach on the LiDAR-MOS benchmark of SemanticKITTI. Our method outperforms the state-of-the-art methods significantly in terms of LiDAR-MOS IoU. Benefiting from the devised coarse-to-fine architecture, our method operates online at sensor frame rate. Code is available at: https://github.com/haomo-ai/MotionSeg3D.

With the field of rigid-body robotics having matured in the last fifty years, routing, planning, and manipulation of deformable objects have recently emerged as a more untouched research area in many fields ranging from surgical robotics to industrial assembly and construction. Routing approaches for deformable objects which rely on learned implicit spatial representations (e.g., Learning-from-Demonstration methods) make them vulnerable to changes in the environment and the specific setup. On the other hand, algorithms that entirely separate the spatial representation of the deformable object from the routing and manipulation, often using a representation approach independent of planning, result in slow planning in high dimensional space. This paper proposes a novel approach to routing deformable one-dimensional objects (e.g., wires, cables, ropes, sutures, threads). This approach utilizes a compact representation for the object, allowing efficient and fast online routing. The spatial representation is based on the geometrical decomposition of the space into convex subspaces, resulting in a discrete coding of the deformable object configuration as a sequence. With such a configuration, the routing problem can be solved using a fast dynamic programming sequence matching method that calculates the next routing move. The proposed method couples the routing and efficient configuration for improved planning time. Our simulation and real experiments show the method correctly computing the next manipulation action in sub-millisecond time and accomplishing various routing and manipulation tasks.

For real-time multirotor kinodynamic planning, the efficiency of sampling-based methods is usually hindered by difficult-to-sample homotopy classes like narrow passages. In this paper, we address this issue by a hybrid scheme. We firstly propose a fast regional optimizer exploiting the information of local environments and then integrate it into a bidirectional global sampling process. The incorporation of the local optimization shows significantly improved success rates and less planning time in various types of challenging environments. We further present a refinement module utilizing the same framework as the regional optimizer. It comprehensively investigates the resulting trajectory of the global sampling and improves its smoothness with nearly negligible computation effort. Benchmark results illustrate that our proposed method can better exploit a previous trajectory compared to the state-of-the-art ones. The planning methods are applied to generate trajectories for a quadrotor system in simulation and real-world, and their capability is validated in real-time applications.

The paper proposes a path planning solution for autonomous robotic exploration of complex subterranean environments. The work contributes to the family of graph-based planners by bringing the following improvements. Firstly, an occupancy grid-based sample-and-project solution to terrain-assessment is proposed instead of building an explicit elevation map of the environment. Secondly, the solution-search method is formulated as a constraint-satisfaction problem to obtain a good-enough solution instead of optimizing for a single objective function parametrized by penalty gains. This method is shown to significantly improve the computational efficiency of the planner. Thirdly, a coordination solution is proposed that relies on the position histories of the robots instead of merged-maps or merged-graphs, therefore, making the planning solution resilient to inter-robot map misalignments, while also reducing the communication bandwidth required to carry out coordination. Finally, the planner also takes into account changes in the environment such as blocked passages that are initially open. The proposed planning solution is demonstrated at the DARPA Subterranean Challenge final event by team MARBLE, the third place finisher of the challenge.

We present a manifold optimization approach to solve inference and planning problems with range constraints. The core of our approach is the definition of a manifold that represents points or poses with range constraints. We discover that the manifold of range-constrained points is homogeneous under the rigid transformation group action, and utilize the group action to derive the tangent space, retraction and topology of the manifold. We evaluate the performance of manifold optimization approach on solving range-constrained inference problems over state-of-the-art constrained optimization methods. The results show that manifold optimization with the range-constraint manifold achieves both faster speed and better constraint satisfaction. We further study the conditions of inference problems that we can treat range measurements as constraints in practice.

In order to be effective general purpose machines in real world environments, robots not only will need to adapt their existing manipulation skills to new circumstances, they will need to acquire entirely new skills on-the-fly. One approach to achieving this capability is via Multi-task Reinforcement Learning (MTRL). Most recent work in MTRL trains a single policy to solve all tasks at once. In this work, we investigate the feasibility of instead training separate policies for each task, and only transferring from a task once the policy for it has finished training. We describe a method of finding near optimal sequences of transfers to perform in this setting, and use it to show that performing the optimal sequence of transfer is competitive with other MTRL methods on the Meta World MT10 benchmark. Lastly, we describe a method for finding nearly optimal transfer sequences during training that is able to improve on training each task from scratch.

Modelling robot dynamics accurately is essential for control, motion optimisation and safe human-robot collaboration. Given the complexity of modern robotic systems, dynamics modelling remains non-trivial, mostly in the presence of compliant actuators, mechanical inaccuracies, friction and sensor noise. Recent efforts have focused on utilising datadriven methods such as Gaussian processes and neural networks to overcome these challenges, as they are capable of capturing these dynamics without requiring extensive knowledge beforehand. While Gaussian processes have shown to be an effective method for learning robotic dynamics with the ability to also represent the uncertainty in the learned model through its variance, they come at a cost of cubic time complexity rather than linear, as is the case for deep neural networks. In this work, we leverage the use of deep kernel models, which combine the computational efficiency of deep learning with the nonparametric flexibility of kernel methods (Gaussian processes), with the overarching goal of realising an accurate probabilistic framework for uncertainty quantification. Through using the predicted variance, we adapt the feedback gains as more accurate models are learned, leading to low-gain control without compromising tracking accuracy. Using simulated and real data recorded from a seven degree-of-freedom robotic manipulator, we illustrate how using stochastic variational inference with deep kernel models increases compliance in the computed torque controller, and retains tracking accuracy. We empirically show how our model outperforms current state-of-the-art methods with prediction uncertainty for online inverse dynamics model learning, and solidify its adaptation and generalisation capabilities across different setups.

For an autonomous vehicle, the ability to sense its surroundings and to build an overall representation of the environment by fusing different sensor data streams is fundamental. To this end, the poses of all sensors need to be accurately determined. Traditional calibration methods are based on: 1) using targets specifically designed for calibration purposes in controlled environments, 2) optimizing a quality metric of the point clouds collected while traversing an unknown but static environment, or 3) optimizing the match among persensor incremental motion observations along a motion path fulfilling special requirements. In real scenarios, however, the online applicability of these methods can be limited, as they are typically highly dynamic, contain degenerate paths, and require fast computations. In this paper, we propose an approach that tackles some of these challenges by formulating the calibration problem as a joint but structured optimization problem of all sensor calibrations that takes as input a summary of the point cloud information consisting of ground points and pole detections. We demonstrate the efficiency and quality of the results of the proposed approach in a set of experiments with LiDAR simulation and real data from an urban trip.

The collective operation of robots, such as unmanned aerial vehicles (UAVs) operating as a team or swarm, is affected by their individual capabilities, which in turn is dependent on their physical design, aka morphology. However, with the exception of a few (albeit ad hoc) evolutionary robotics methods, there has been very little work on understanding the interplay of morphology and collective behavior. There is especially a lack of computational frameworks to concurrently search for the robot morphology and the hyper-parameters of their behavior model that jointly optimize the collective (team) performance. To address this gap, this paper proposes a new co-design framework. Here the exploding computational cost of an otherwise nested morphology/behavior co-design is effectively alleviated through the novel concept of “talent” metrics; while also allowing significantly better solutions compared to the typically sub-optimal sequential morphology → behavior design approach. This framework comprises four major steps: talent metrics selection, talent Pareto exploration (a multi-objective morphology optimization process), behavior optimization, and morphology finalization. This co-design concept is demonstrated by applying it to design UAVs that operate as a team to localize signal sources, e.g., in victim search and hazard localization. Here, the collective behavior is driven by a recently reported batch Bayesian search algorithm called Bayes-Swarm. Our case studies show that the outcome of co-design provides significantly higher success rates in signal source localization compared to a baseline design, across a variety of signal environments and teams with 6 to 15 UAVs. Moreover, this co-design process provides two orders of magnitude reduction in computing time compared to a projected nested design approach.