ICRA 2024 Papers

In this paper, we introduce a double-stage transfer learning framework based on expert data. It employs probabilistic graphical models to effectively capture helical periodic features in the latent space, integrating Bayesian variational inference and neural networks for implementation. Compared to traditional methods, it achieves high precision and stable control even in environments with limited observation signals and high noise levels. We have successfully applied this method to a biomedical task of a simulated cranial window procedure. Preliminary results show promising performance comparable to those of human experts with only image information, further validating the efficacy of the proposed method.

Although pre-training on a large amount of data is beneficial for robot learning, current paradigms only perform large-scale pretraining for visual representations, whereas representations for other modalities are trained from scratch. In contrast to the abundance of visual data, it is unclear what relevant internet-scale data may be used for pretraining other modalities such as tactile sensing. Such pretraining becomes increasingly crucial in the low-data regimes common in robotics applications. In this paper, we address this gap by using contact microphones as an alternative tactile sensor. Our key insight is that contact microphones capture inherently audio-based information, allowing us to leverage large-scale audio-visual pretraining to obtain representations that boost the performance of robotic manipulation. To the best of our knowledge, our method is the first approach leveraging large-scale multisensory pre-training for robotic manipulation. For supplementary information including videos of real robot experiments, please see https://sites.google.com/view/hearing-touch.

In this work, we delve into the intricate synergy among non-prehensile actions like pushing, and prehensile actions such as grasping and throwing, within the domain of robotic manipulation. We introduce an innovative approach to learning these synergies by leveraging model-free deep reinforcement learning. The robot’s workflow involves detecting the pose of the target object and the basket at each time step, predicting the optimal push configuration to isolate the target object, determining the appropriate grasp configuration, and inferring the necessary parameters for an accurate throw into the basket. This empowers robots to skillfully reconfigure cluttered scenarios through pushing, creating space for collision-free grasping actions. Simultaneously, we integrate throwing behavior, showcasing how this action significantly extends the robot’s operational reach. Ensuring safety, we developed a simulation environment in Gazebo for robot training, applying the learned policy directly to our real robot. Notably, this work represents a pioneering effort to learn the synergy between pushing, grasping, and throwing actions. Extensive experimentation in both simulated and real-robot scenarios substantiates the effectiveness of our approach across diverse settings. Our approach achieves a success rate exceeding 80% in both simulated and real-world scenarios. A video showcasing our experiments is available online at: https://youtu.be/q1l4BJVDbRw

In this paper, the dynamics of an emerging class of rotating nature-inspired micro aerial vehicles known as the Monocopter is proven and shown to be differentially flat. By exploiting this phenomenon, trajectory tracking can now be implemented on Monocopters via feed-forward terms that are computed per the trajectory. To demonstrate this, a Monocopter in the form of a Stable Invertible Coaxial Actuated ROtorcraft (SICARO) is chosen to harness this approach fully. The SICARO is capable of flying with either side of the wing facing up and this feature determines the craft’s direction of rotation about its body Z axis as well. In addition, it has the unique feature of a coaxial motor configuration that allows for a pitching-up moment regardless of the wing side facing up. The feed-forward terms computed are fused into a cascaded nonlinear controller on the craft to ensure its effectiveness in tracking trajectories. Lastly, the flight experiments extend to both sides of the wing to validate this method as being applicable for trajectory tracking for Monocopters such as the SICARO which has an extended range of flying capabilities.

Deep-sea research represents invaluable opportunities to unravel hidden ecosystems, uncover unknown biodiversity, and provide critical insights into the Earth’s history and the impacts of climate change. Due to the extreme conditions, exploring the deep-sea traditionally requires costly equipment, such as specific diving robots, engineered to withstand the high pressure. Our research aims to reduce the costs of deep-sea sediment sampling by introducing a novel actuation system for suction samplers, that capitalises the advantages of soft material actuators. At first glance, soft material actuators may not appear suitable for the harsh conditions that prevail in the deep-sea, but when combined with a rigid, bistable mechanism there is great potential for improving the accessibility of sampling and research in this challenging environment. The binary actuation system that results from this combination, is modular, scalable, lightweight, and low cost in comparison to existing solutions.

Skull base surgery is a demanding field in which surgeons operate in and around the skull while avoiding critical anatomical structures including nerves and vasculature. While image-guided surgical navigation is the prevailing standard, limitation still exists requiring personalized planning and recognizing the irreplaceable role of a skilled surgeon. This paper presents a collaboratively controlled robotic system tailored for assisted drilling in skull base surgery. Our central hypothesis posits that this collaborative system, enriched with haptic assistive modes to enforce virtual fixtures, holds the potential to significantly enhance surgical safety, streamline efficiency, and alleviate the physical demands on the surgeon. The paper describes the intricate system development work required to enable these virtual fixtures through haptic assistive modes. To validate our system’s performance and effectiveness, we conducted initial feasibility experiments involving a medical student and two experienced surgeons. The experiment focused on drilling around critical structures following cortical mastoidectomy, utilizing dental stone phantom and cadaveric models. Our experimental results demonstrate that our proposed haptic feedback mechanism enhances the safety of drilling around critical structures compared to systems lacking haptic assistance. With the aid of our system, surgeons were able to safely skeletonize the critical structures without breaching any critical structure even under obstructed view of the surgical site.

Various heuristic objectives for modeling hand-object interaction have been proposed in past work. However, due to the lack of a cohesive framework, these objectives often possess a narrow scope of applicability and are limited by their efficiency or accuracy. In this paper, we propose HANDYPRIORS, a unified and general pipeline for pose estimation in human-object interaction scenes by leveraging recent advances in differentiable physics and rendering. Our approach employs rendering priors to align with input images and segmentation masks along with physics priors to mitigate penetration and relative-sliding across frames. Furthermore, we present two alternatives for hand and object pose estimation. The optimization-based pose estimation achieves higher accuracy, while the filtering-based tracking, which utilizes the differentiable priors as dynamics and observation models, executes faster. We demonstrate that HANDYPRIORS attains comparable or superior results in the pose estimation task, and that the differentiable physics module can predict contact information for pose refinement. We also show that our approach generalizes to perception tasks, including robotic hand manipulation and human-object pose estimation in the wild.

This paper presents a method to learn hand-object interaction prior for reconstructing a 3D hand-object scene from a single RGB image. The inference as well as training-data generation for 3D hand-object scene reconstruction is challenging due to the depth ambiguity of a single image and occlusions by the hand and object. We turn this challenge into an opportunity by utilizing the hand shape to constrain the possible relative configuration of the hand and object geometry. We design a generalizable implicit function, HandNeRF, that explicitly encodes the correlation of the 3D hand shape features and 2D object features to predict the hand and object scene geometry. With experiments on real-world datasets, we show that HandNeRF can reconstruct hand-object scenes of novel grasp configurations more accurately than comparable methods. Moreover, we demonstrate that object reconstruction from HandNeRF ensures more accurate execution of downstream tasks, such as grasping for robotic hand-over.

Reliable autonomous navigation requires adapting the control policy of a mobile robot in response to dynamics changes in different operational conditions. Hand-designed dynamics models may struggle to capture model variations due to a limited set of parameters. Data-driven dynamics learning approaches offer higher model capacity and better generalization but require large amounts of state-labeled data. This paper develops an approach for learning robot dynamics directly from point-cloud observations, removing the need and associated errors of state estimation, while embedding Hamiltonian structure in the dynamics model to improve data efficiency. We design an observation-space loss that relates motion prediction from the dynamics model with motion prediction from point-cloud registration to train a Hamiltonian neural ordinary differential equation. The learned Hamiltonian model enables the design of an energy-shaping model-based tracking controller for rigid-body robots. We demonstrate dynamics learning and tracking control on a real nonholonomic wheeled robot.

The ability of a robot to perceive and understand its environment is crucial for its actions and behavior. Humans are adept at using semantic information for object localization and path planning, a skill that robots need to emulate for intelligent adaptation in dynamic settings. Training of the spatial anticipation ability, which can enhance spatial perception through semantic understanding, necessitates the availability of appropriate data. Although extensive research has been conducted on datasets for outdoor environments, especially in the context of autonomous driving, there is still a notable lack of datasets specifically designed for indoor environments, with a focus on dynamic object localization. This paper introduces HabitatDyn 2.0, a dataset specifically designed for enhancing object localization capabilities with semantic information from a robot’s perspective. Besides RGB videos, semantic annotations, and depth information, HabitatDyn 2.0 also features top-down view labels for dynamic objects, which is required for training the spatial anticipation ability based on semantic information. Additionally, an algorithm that leverages spatial anticipation for dynamic object localization is presented, trained, and evaluated on the dataset.

Visual Semantic Navigation (VSN) aims at navigating a robot to a given target object in a previously unseen scene. To tackle this task, the robot must learn a nimble navigation policy by utilizing spatial patterns and semantic co-occurrence relations among objects in the scene. Prevailing approaches extract scene priors from the instant visual observations and solidify them in neural episodic memory to achieve flexible navigation. However, due to the oblivion and underuse of the scene priors, these methods are plagued by repeated exploration, effective-knowledge sparsity, and wrong decisions. To alleviate these issues, we propose a novel VSN policy, HSPNav, based on Hierarchical Scene Priors (HSP) and Deep Reinforcement Learning (DRL). The HSP contains two components, i.e., the egocentric semantic map-based Local Scene Priors (LSP) and the commonsense relational graph-based Global Scene Priors (GSP). Then, efficient semantic navigation is achieved by employing an immediate LSP to retrieve conducive contextual memories from the GSP. By utilizing the MP3D dataset, the experimental results in the Habitat simulator demonstrate that our HSP brings a significant boost over the baselines. Furthermore, we take an essential step from simulation to reality by bridging the gap from Habitat to ROS. The migration evaluations show that HSPNav can generalize to realistic settings well and achieve promising performance.

Absolute Pose Regressors (APRs) directly estimate camera poses from monocular images, but their accuracy is unstable for different queries. Uncertainty-aware APRs provide uncertainty information on the estimated pose, alleviating the impact of these unreliable predictions. However, existing uncertainty modelling techniques are often coupled with a specific APR architecture, resulting in suboptimal performance compared to state-of-the-art (SOTA) APR methods. This work introduces a novel APR-agnostic framework, HR-APR, that formulates uncertainty estimation as cosine similarity estimation between the query and database features. It does not rely on or affect APR network architecture, which is flexible and computationally efficient. In addition, we take advantage of the uncertainty for pose refinement to enhance the performance of APR. The extensive experiments demonstrate the effectiveness of our framework, reducing 27.4% and 15.2% of computational overhead on the 7Scenes and Cambridge Landmarks datasets while maintaining the SOTA accuracy in single-image APRs.

To develop the next generation of intelligent LiDARs, we propose a novel framework of parallel LiDARs and construct a hardware prototype in our experimental platform, DAWN (Digital Artificial World for Natural). It emphasizes the tight integration of physical and digital space in LiDAR systems, with networking being one of its supported core features. In the context of autonomous driving, V2V (Vehicle-to-Vehicle) technology enables efficient information sharing between different agents which significantly promotes the development of LiDAR networks. However, current research operates under an ideal situation where all vehicles are equipped with identical LiDAR, ignoring the diversity of LiDAR categories and operating frequencies. In this paper, we first utilize OpenCDA and RLS (Realistic LiDAR Simulation) to construct a novel heterogeneous LiDAR dataset named OPV2V-HPL. Additionally, we present HPL-ViT, a pioneering architecture designed for robust feature fusion in heterogeneous and dynamic scenarios. It uses a graph-attention Transformer to extract domain-specific features for each agent, coupled with a cross-attention mechanism for the final fusion. Extensive experiments on OPV2V-HPL demonstrate that HPL-ViT achieves SOTA (state-of-the-art) performance in all settings and exhibits outstanding generalization capabilities.

Visual SLAM is an essential tool in diverse applications such as robot perception and extended reality, where feature-based methods are prevalent due to their accuracy and robustness. However, existing methods employ either hand-crafted or solely learnable point features and are thus limited by the feature attributes. In this paper, we propose incorporating hybrid point features efficiently into a single system. By integrating hand-crafted and learnable features, we seek to capitalize on their complementary attributes in both key-point identification and descriptor expressiveness. To this purpose, we design a pre-processing module, which includes extraction, inter-class processing, and post-processing of hybrid point features. We present an efficient matching approach to exclusively perform the data association within the same class of features. Moreover, we design a Hybrid Bag-of-Words (H-BoW) model to deal with hybrid point features in matching and loop-closure-detection. By integrating the proposed framework into a modern feature-based system, we introduce HPF-SLAM. We evaluate the system on EuRoC-MAV and TUM-RGBD benchmarks. The experimental results show that our method consistently surpasses the baseline at comparable speed.

A good representation of a large, complex mobile robot workspace must be space-efficient yet capable of encoding relevant geometric details. When exploring unknown environments, it needs to be updatable incrementally in an online fashion. We introduce HIO-SDF, a new method that represents the environment as a Signed Distance Field (SDF). State of the art representations of SDFs are based on either neural networks or voxel grids. Neural networks are capable of representing the SDF continuously. However, they are hard to update incrementally as neural networks tend to forget previously observed parts of the environment unless an extensive sensor history is stored for training. Voxel-based representations do not have this problem but they are not space-efficient especially in large environments with fine details. HIO-SDF combines the advantages of these representations using a hierarchical approach which employs a coarse voxel grid that captures the observed parts of the environment together with high-resolution local information to train a neural network. HIO-SDF achieves a 46% lower mean global SDF error across all test scenes than a state of the art continuous representation, and a 30% lower error than a discrete representation at the same resolution as our coarse global SDF grid. Videos and code are available at: https://samsunglabs.github.io/HIO-SDF-project-page/

Small object detection has been a challenging problem in the field of object detection. There has been some works that proposes improvements for this task, such as adding several attention blocks or changing the whole structure of feature fusion networks. However, the computation cost of these models is large, which makes deploying a real-time object detection system unfeasible, while leaving room for improvement. To this end, an improved YOLOv5 model: HIC-YOLOv5 is proposed to address the aforementioned problems. Firstly, an additional prediction head specific to small objects is added to provide a higher-resolution feature map for better prediction. Secondly, an involution block is adopted between the backbone and neck to increase channel information of the feature map. Moreover, an attention mechanism named CBAM is applied at the end of the backbone, thus not only decreasing the computation cost compared with previous works but also emphasizing the important information in both channel and spatial domain. Our result shows that HIC-YOLOv5 has improved mAP@[.5:.95] by 6.42% and mAP@0.5 by 9.38% on VisDrone-2019-DET dataset.

In many intelligent transportation systems, predicting the future motion of heterogeneous traffic participants is a fundamental but challenging task due to various factors encompassing the agents’ dynamic states, interactions with neighboring agents and surrounding traffic infrastructures, and their stochastic and multi-modal natural behavior tendencies. However, existing approaches have limitations as they either focus solely on static, pairwise interactions, ignoring interactions of varied granularity, or fail to tackle agents’ heterogeneity. In this paper, instead of focusing solely on pairwise interactions, we propose a Heterogenous Hypergraph Graph Neural Network (HHGNN) based motion prediction model that leverages the nature of hypergraph to encode the groupwise interactions among traffic participants. Moreover, we propose the type-aware two-level hypergraph message passing module (TTHMS) with learnable hyperedge-type embeddings to model the intra-group and inter-group level interactions among heterogeneous traffic agents (e.g., vehicles, pedestrians, and cyclists). Besides, We integrate a scene context fusion layer in TTHMS to incorporate the scene context. Comparison and ablation experiments on the Waymo Open Motion Dataset (WOMD) demonstrate HHGNN’s effectiveness within the motion prediction task.

Simultaneous Localization and Mapping (SLAM) is a fundamental task in robotics, driving numerous applications such as autonomous driving and virtual reality. Recent progress on neural implicit SLAM has shown encouraging and impressive results. However, the robustness of neural SLAM, particularly in challenging or data-limited situations, remains an unresolved issue. This paper presents HERO-SLAM, a Hybrid Enhanced Robust Optimization method for neural SLAM, which combines the benefits of neural implicit field and feature-metric optimization. This hybrid method optimizes a multi-resolution implicit field and enhances robustness in challenging environments with sudden viewpoint changes or sparse data collection. Our comprehensive experimental results on benchmarking datasets validate the effectiveness of our hybrid approach, demonstrating its superior performance over existing implicit field-based methods in challenging scenarios. HERO-SLAM provides a new pathway to enhance the stability, performance, and applicability of neural SLAM in real-world scenarios. Project page: https://hero-slam.github.io.

Given its wide application in robotics, point cloud registration is a widely researched topic. Conventional methods aim to find a rotation and translation that align two point clouds in 6 degrees of freedom (DoF). However, certain tasks in robotics, such as category-level pose estimation, involve non-uniformly scaled point clouds, requiring a 9DoF transform for accurate alignment. We propose HEGN, a novel equivariant graph neural network for 9DoF point cloud registration. HEGN utilizes equivariance to rotation, translation, and scaling to estimate the transformation without relying on point correspondences. Based on graph representations for both point clouds, we extract equivariant node features aggregated in their local, cross-, and global context. In addition, we introduce a novel node pooling mechanism that leverages the cross-context importance of nodes to pool the graph representation. By repeating the feature extraction and node pooling, we obtain a graph hierarchy. Finally, we determine rotation and translation by aligning equivariant features aggregated over the graph hierarchy. To estimate scaling, we leverage scale information in the vector norm of the equivariant features. We evaluate the effectiveness of HEGN through experiments with the synthetic ModelNet40 dataset and the real-world ScanObjectNN dataset. The results show the superior performance of HEGN in 9DoF point cloud registration and its competitive performance in conventional 6DoF point cloud registration.