IROS 2022 Papers

Electroadhesion is a versatile and controllable adhesion mechanism that has been used extensively in robotics. Soft electroadhesion embodies electrostatic adhesion in soft materials and is required for shape-adaptive and safe grasping of curved objects and delicate materials. In this work, we present a soft electroadhesive fabrication method based on laser scribing graphene oxide on a silicone film, which is cost-effective, facile and green. The method can be used to generate complex electroadhesive patterns without molds or stencils. We then present a 2D finite element model to demonstrate the shape-changing behavior and electric field distributions of a dual-mode parallel dielectric elastomer actuation and coplanar electroadhesion structure. The soft electroadhesive fabrication method based on laser-scribed graphene oxide electrodes and its experimental characterization results, together with its shape-morphing simulation model are expected to enable the wider adoption of soft electroadhesion in future robotics.

Navigating off-road with a fast autonomous vehicle depends on a robust perception system that differentiates traversable from non-traversable terrain. Typically, this depends on a semantic understanding which is based on supervised learning from images annotated by a human expert. This requires a significant investment in human time, assumes correct expert classification, and small details can lead to misclassification. To address these challenges, we propose a method for predicting high- and low-risk terrains from only past vehicle experience in a self-supervised fashion. First, we develop a tool that projects the vehicle trajectory into the front camera image. Second, occlusions in the 3D representation of the terrain are filtered out. Third, an autoencoder trained on masked vehicle trajectory regions identifies low- and high-risk terrains based on the reconstruction error. We evaluated our approach with two models and different bottleneck sizes with two different training and testing sites with a four-wheeled off-road vehicle. Comparison with two independent test sets of semantic labels from similar terrain as training sites demonstrates the ability to separate the ground as low-risk and the vegetation as high-risk with 81.1% and 85.1% accuracy.

Domain adaptation is an important property in robot vision, which enables the neural networks pre-trained on source domains to adapt target domains automatically without any annotation efforts. During this process, source data is not always accessible due to the constraints of expensive storage overhead and data privacy protection. Therefore, the source domain pre-trained model is expected to optimize with only unlabeled target data, termed as source-free unsupervised domain adaptation. In this paper, we view this problem as a special case of noisy label learning, since the given pre-trained model can generate noisy labels for unlabeled target data via network inference. The potential semantic cues for unsupervised domain adaptation exactly lie on these noisy labels. Inspired by this problem modeling, we propose a simple yet effective Self-Supervised Noisy Label Learning method, which injects self-supervised learning to impose the intrinsic data structure and facilitate label-denoising. Extensive experiments have been conducted on diverse benchmarks to validate the effectiveness. Our method achieves state-of-the-art performance.

Self-supervised learning on point clouds has gained a lot of attention recently, since it addresses the label-efficiency and domain-gap problems on point cloud tasks. In this paper, we propose a novel self-supervised framework to learn informative features from partial point clouds. We leverage partial point clouds scanned by LiDAR that contain both content and pose attributes, and we show that disentangling such two factors from partial point clouds enhances feature learning. To this end, our framework consists of three main parts: 1) a completion network to capture holistic semantics of point clouds; 2) a pose regression network to understand the viewing angle where partial data is scanned from; 3) a partial reconstruction network to encourage the model to learn content and pose features. To demonstrate the robustness of the learnt feature representations, we conduct several downstream tasks including classification, part segmentation, and registration, with comparisons against state-of-the-art methods. Our method not only outperforms existing self-supervised methods, but also shows a better generalizability across synthetic and real-world datasets.

There are many spaces inaccessible to humans where robots could help deliver sensors and equipment. Many of these spaces contain three-dimensional passageways and uneven terrain that pose challenges for robot design and control. Everting toroidal robots, which move via simultaneous eversion and inversion of their body material, are promising for navigation in these types of spaces. We present a novel soft everting toroidal robot that propels itself using a motorized device inside an air-filled membrane. Our robot requires only a single control signal to move, can conform to its environment, and can climb vertically with a motor torque that is independent of the force used to brace the robot against its environment. We derive and validate models of the forces involved in its motion, and we demonstrate the robot's ability to navigate a maze and climb a pipe.

Multiple object tracking in 3D point clouds has applications in mobile robots and autonomous driving. This is a challenging problem due to the sparse nature of the point clouds and the added difficulty of annotation in 3D for supervised learning. To overcome these challenges, we propose a neural network architecture that learns effective object features and their affinities in a self supervised fashion for multiple object tracking in 3D point clouds captured with LiDAR sensors. For self supervision, we use two approaches. First, we generate two augmented LiDAR frames from a single real frame by applying translation, rotation and cutout to the objects. Second, we synthesize a LiDAR frame using CAD models or primitive geometric shapes and then apply the above three augmentations to them. Hence, the ground truth object locations and associations are known in both frames for self supervision. This removes the need to annotate object associations in real data, and additionally the need for training data collection and annotation for object detection in synthetic data. To the best of our knowledge, this is the first self supervised multiple object tracking method for 3D data. Our model achieves state of the art results.

This paper introduces a method to generate highly selective encodings that can be magnetically “programmed” onto physical modules to enable them to self-assemble in chosen configurations. We generate these encodings based on Hadamard matrices, and show how to design the faces of modules to be maximally attractive to their intended mate, while remaining maximally agnostic to other faces. We derive guarantees on these bounds, and verify their attraction and agnosticism experimentally. Using cubic modules whose faces have been covered in soft magnetic material, we show how inexpensive, passive modules with planar faces can be used to selectively self-assemble into target shapes without geometric guides. We show that these modules can be easily re-programmed for new target shapes using a CNC-based magnetic plotter, and demonstrate self-assembly of 8 cubes in a water tank.

Markov decision processes (MDPs) are a common general-purpose model used in robotics for representing sequential decision-making problems. Given the complexity of robotics applications, a popular approach for approximately solving MDPs relies on state aggregation to reduce the size of the state space but at the expense of policy fidelity-offering a trade-off between policy quality and computation time. Naturally, this poses a challenging metareasoning problem: how can an autonomous system dynamically select different state abstractions that optimize this trade-off as it operates online? In this paper, we formalize this metareasoning problem with a notion of time-dependent utility and solve it using deep reinforcement learning. To do this, we develop several general, cheap heuristics that summarize the reward structure and transition topology of the MDP at hand to serve as effective features. Empirically, we demonstrate that our metareasoning approach outperforms several baseline approaches and a strong heuristic approach on a standard benchmark domain.

Place recognition is seen as a crucial factor to correct cumulative errors in Simultaneous Localization and Mapping (SLAM) applications. Most existing studies focus on visual place recognition, which is inherently sensitive to environmental changes such as illumination, weather and seasons. Considering these facts, more recent attention has been attracted to use 3-D Light Detection and Ranging (LiDAR) scans for place recognition, which demonstrates more credibility by exerting accurate geometric information. Different from pure geometric-based studies, this paper proposes a novel global descriptor, named SectionKey, which leverages both semantic and geometric information to tackle the problem of place recognition in large-scale urban environments. The proposed descriptor is robust and invariant to viewpoint changes. Specifically, the encoded three-layers key serves as a pre-selection step and a ‘candidate center’ selection strategy is deployed before calculating the similarity score, thus improving the accuracy and efficiency significantly. Then, a two-step semantic iterative closest point (ICP) algorithm is applied to acquire the 3-D pose (x, y, θ) that is used to align the candidate point clouds with the query frame and calculate the similarity score. Extensive experiments have been conducted on public Semantic KITTI dataset to demonstrate the superior performance of our proposed system over state-of-the-art baselines.

We present the first real-time system capable of tracking and reconstructing, individually, every visible object in a given scene, without any form of prior on the rigidness of the objects, texture existence, or object category. In contrast with previous methods such as Co-Fusion and MaskFusion that first segment the scene into individual objects and then process each object independently, the proposed method dynamically segments the non-rigid scene as part of the tracking and reconstruction process. When new measurements indicate topology change, reconstructed models are updated in real-time to reflect that change. Our proposed system can provide the live geometry and deformation of all visible objects in a novel scene in real-time, which makes it possible to be integrated seamlessly into numerous existing robotics applications that rely on object models for grasping and manipulation. The capabilities of the proposed system are demonstrated in challenging scenes that contain multiple rigid and non-rigid objects. Supplementary material, including video, can be found at https://github.com/changhaonan/STAR-no-prior.

Studies in robot teleoperation have been centered around action specifications-from continuous joint control to discrete end-effector pose control. However, these “robot-centric” interfaces often require skilled operators with extensive robotics expertise. To make teleoperation accessible to nonexpert users, we propose the framework “Scene Editing as Teleoperation” (SEaT), where the key idea is to transform the traditional “robot-centric” interface into a “scene-centric” interface-instead of controlling the robot, users focus on specifying the task's goal by manipulating digital twins of the real-world objects. As a result, a user can perform teleoperation without any expert knowledge of the robot hardware. To achieve this goal, we utilize a category-agnostic scene-completion algorithm that translates the real-world workspace (with unknown objects) into a manipulable virtual scene representation and an action-snapping algorithm that refines the user input before generating the robot's action plan. To train the algorithms, we procedurely generated a large-scale, diverse kit-assembly dataset that contains object-kit pairs that mimic real-world object-kitting tasks. Our experiments in simulation and on a real-world system demonstrate that our framework improves both the efficiency and success rate for 6DoF kit-assembly tasks. A user study demonstrates that SEaT framework participants achieve a higher task success rate and report a lower subjective workload compared to an alternative robot-centric interface.

We present a generic framework for scale-aware direct monocular odometry based on depth prediction from a deep neural network. In contrast with previous methods where depth information is only partially exploited, we formulate a novel depth prediction residual which allows us to incorporate multi-view depth information. In addition, we propose to use a truncated robust cost function which prevents considering inconsistent depth estimations. The photometric and depth-prediction measurements are integrated into a tightly-coupled optimization leading to a scale-aware monocular system which does not accumulate scale drift. Our proposal does not particularize for a concrete neural network, being able to work along with the vast majority of the existing depth prediction solutions. We demonstrate the validity and generality of our proposal evaluating it on the KITTI odometry dataset, using two publicly available neural networks and comparing it with similar approaches and the state-of-the-art for monocular and stereo SLAM. Experiments show that our proposal largely outperforms classic monocular SLAM, being 5 to 9 times more precise, beating similar approaches and having an accuracy which is closer to that of stereo systems.

The scale ambiguity problem is inherently unsolvable to monocular SLAM without the metric baseline between moving cameras. In this paper, we present a novel scale estimation approach based on an object-level SLAM system. To obtain the absolute scale of the reconstructed map, we formulate an optimization problem to make the scaled dimensions of objects conform to the distribution of their sizes in the physical world, without relying on any prior information about gravity direction. The dual quadric is adopted to represent objects for its ability to describe objects compactly and accurately, thus providing reliable dimensions for scale estimation. In the proposed monocular object-level SLAM system, semantic objects are initialized first from fitted 3-D oriented bounding boxes and then further optimized under constraints of 2-D detections and 3-D map points. Experiments on indoor and outdoor public datasets show that our approach outperforms existing methods in terms of accuracy and robustness.

This paper advocates the Gaussian belief propagation solver for factor graphs in the case of gas distribution mapping to support an olfactory sensing robot. The local message passing of belief propagation moves away from the standard Cholesky decomposition technique, which avoids solving the entire factor graph at once and allows for only areas of interest to be updated more effectively. Implementing a local solver means that iterative updates to the distribution map can be achieved orders of magnitude quicker than conventional direct solvers which scale computationally to the size of the map. After defining the belief propagation algorithm for gas mapping, several state of the art message scheduling algorithms are tested in simulation against the standard Cholesky solver for their ability to converge to the exact solution. Testing shows that under the wildfire scheduling method for a large urban scenario, that distribution maps can be iterated at least 10 times faster whilst still maintaining exact solutions. This move to an efficient local framework allows future works to consider 3D mapping, predictive utility and multi-robot distributed mapping.

Navigating accurately in potentially GPS-denied environments is a perquisite of autonomous systems. Relative localization based on ultra-wideband (UWB) is - especially indoors - a promising technology. In this paper, we present a probabilistic filter based Modular Multi-Sensor Fusion (MMSF) approach with the capability of using efficiently all information in a fully meshed UWB ranging network. This allows an accurate mobile agent state estimation and the calibration of the ranging network's spatial constellation. We advocate a new paradigm that includes elements from Collaborative State Estimation (CSE) and allows us considering all stationary UWB anchors and the mobile agent as a decentralized set of estimtors/filters. With this, our method can include all meshed (inter-)sensor observations tightly coupled in a modular estimator. We show that the application of our CSE-inspired method in such a context breaks the computational barrier. Otherwise, it would, for the sakeof complexity-reduction, prohibit the use of all available information or would lead to significant estimator inconsistencies due to coarse approximations. We compare the proposed approach against different MMSF strategies in terms of execution time, accuracy, and filter credibility on both synthetic data and on a dataset from real Unmanned Aerial Vehicles (UAVs).

Evaluating rare but high-stakes events is one of the main challenges in obtaining reliable reinforcement learning policies, especially in large or infinite state/action spaces where limited scalability dictates a prohibitively large number of testing iterations. On the other hand, a biased or inaccurate policy evaluation in a safety-critical system could potentially cause unexpected catastrophic failures during deployment. This paper proposes the Accelerated Policy Evaluation (APE) method, which simultaneously uncovers rare events and estimates the rare event probability in Markov decision processes. The APE method treats the environment nature as an adversarial agent and learns towards, through adaptive importance sampling, the zero-variance sampling distribution for the policy evaluation. Moreover, APE is scalable to large discrete or continuous spaces by incorporating function approximators. We investigate the convergence property of APE in the tabular setting. Our empirical studies show that APE can estimate the rare event probability with a smaller bias while only using orders of magnitude fewer samples than baselines in multi-agent and single-agent environments.

Online coverage path planning to explore an unknown workspace with multiple homogeneous robots could be either centralized or distributed. While distributed planners are computationally faster, centralized planners can produce more efficient paths, reducing the duration of completing a coverage mission significantly. To exploit the power of a centralized framework, we propose a receding horizon centralized online multi-robot planner. In each planning horizon, it generates collision-free paths that guide the robots to visit some obstacle-free locations (aka goals) not visited so far, which in turn help them explore some new regions with their laser rangefinders. We formally prove that, under reasonable conditions, it enables the robots to cover a workspace completely and subsequently analyze its time complexity. We evaluate our planner for ground and aerial robots by performing experiments with up to 128 robots on six 2D grid-based benchmark obstacle maps, establishing scalability. We also perform Gazebo simulations with 10 quadcopters and real experiments with 2 four-wheel ground robots, demonstrating its practical feasibility. Further-more, a comparison with a state-of-the-art distributed planner establishes its superiority in coverage completion time.

Reinforcement learning algorithms require a large amount of samples; this often limits their real-world applications on even simple tasks. Such a challenge is more outstanding in multi-agent tasks, as each step of operation is more costly, requiring communications or shifting or resources. This work aims to improve data efficiency of multi-agent control by model-based learning. We consider networked systems where agents are cooperative and communicate only locally with their neighbors, and propose the decentralized model-based policy optimization framework (DMPO). In our method, each agent learns a dynamic model to predict future states and broadcast their predictions by communication, and then the policies are trained under the model rollouts. To alleviate the bias of model-generated data, we restrain the model usage for generating myopic rollouts, thus reducing the compounding error of model generation. To pertain the independence of policy update, we introduce extended value function and theoretically prove that the resulting policy gradient is a close approximation to true policy gradients. We evaluate our algorithm on several benchmarks for intelligent transportation systems, which are connected autonomous vehicle control tasks (Flow and CACC) and adaptive traffic signal control (ATSC). Empirical results show that our method achieves superior data efficiency and matches the performance of model-free methods using true models. The source code of our algorithm and baselines can be found at https://github.com/PKU-MARL/Model-Based-MARL.

This paper presents an accurate and scalable method for fiducial tag localization on a 3D prior environmental map. The proposed method comprises three steps: 1) visual odometry-based landmark SLAM for estimating the relative poses between fiducial tags, 2) geometrical matching-based global tag-map registration via maximum clique finding, and 3) tag pose refinement based on direct camera-map alignment with normalized information distance. Through simulation-based evaluations, the proposed method achieved a 98 % global tag-map registration success rate and an average tag pose estimation accuracy of a few centimeters. Experimental results in a real environment demonstrated that it enables to localize over 110 fiducial tags placed in an environment in 25 minutes for data recording and post-processing.