ICRA 2024 论文列表

Existing grasp prediction approaches are mostly based on offline learning, while, ignoring the exploratory grasp learning during online adaptation to new picking scenarios, i.e., objects that are unseen or out-of-domain (OOD), camera and bin settings, etc. In this paper, we present an uncertainty-based approach for online learning of grasp predictions for robotic bin picking. Specifically, the online learning algorithm with an effective exploration strategy can significantly improve its adaptation performance to unseen environment settings. To this end, we first propose to formulate online grasp learning as an RL problem that will allow us to adapt both grasp reward prediction and grasp poses. We propose various uncertainty estimation schemes based on Bayesian uncertainty quantification and distributional ensembles. We carry out evaluations on real-world bin picking scenes of varying difficulty. The objects in the bin have various challenging physical and perceptual characteristics that can be characterized by semi- or total transparency, and irregular or curved surfaces. The results of our experiments demonstrate a notable improvement of grasp performance in comparison to conventional online learning methods which incorporate only naive exploration strategies. Video: https://youtu.be/fPKOrjC2QrU

Fleets of unmanned robots can be beneficial for the long-term monitoring of large areas, e.g., to monitor wild flocks, detect intruders, search and rescue. Monitoring numerous dynamic targets in a collaborative and efficient way is a challenging problem that requires online coordination and information fusion. The majority of existing works either assume a passive all-to-all observation model to minimize the summed uncertainties over all targets by all robots, or optimize over the jointed discrete actions while neglecting the dynamic constraints of the robots and unknown behaviors of the targets. This work proposes an online task and motion coordination algorithm that ensures an explicitly-bounded estimation uncertainty for the target states, while minimizing the average number of active robots. The robots have a limited-range perception to actively track a limited number of targets simultaneously, of which their future control decisions are all unknown. It includes: (i) the assignment of monitoring tasks, modeled as a flexible size multiple vehicle routing problem with time windows (m-MVRPTW), given the predicted target trajectories with uncertainty measure in the road-networks; (ii) the nonlinear model predictive control (NMPC) for optimizing the robot trajectories under uncertainty and safety constraints. It is shown that the robots can switch between active and inactive roles dynamically online as required by the unknown monitoring task. The proposed methods are validated via large-scale simulations of up to 100 robots and targets.

In this paper, we investigate a hybrid scheme that combines nonlinear model predictive control (MPC) and model-based reinforcement learning (RL) for navigation planning of an autonomous model car across offroad, unstructured terrains without relying on predefined maps. Our innovative approach takes inspiration from BADGR, an LSTM-based network that primarily concentrates on environment modeling, but distinguishes itself by substituting LSTM modules with transformers to greatly elevate the performance of our model. Addressing uncertainty within the system, we train an ensemble of predictive models and estimate the mutual information between model weights and outputs, facilitating dynamic horizon planning through the introduction of variable speeds. Further enhancing our methodology, we incorporate a nonlinear MPC controller that accounts for the intricacies of the vehicle’s model and states. The model-based RL facet produces steering angles and quantifies inherent uncertainty. At the same time, the nonlinear MPC suggests optimal throttle settings, striking a balance between goal attainment speed and managing model uncertainty influenced by velocity. In the conducted studies, our approach excels over the existing baseline by consistently achieving higher metric values in predicting future events and seamlessly integrating the vehicle’s kinematic model for enhanced decision-making. The code and the evaluation data are available at (Github-repo).

While existing Learning from intervention (LfI) methods within the human-in-the-loop reinforcement learning (HiL-RL) paradigm mainly operate on the assumption that human policies are homogeneous and deterministic with low variance, natural human driving behaviors are multimodal with intrinsic uncertainties, and hence, accommodating diverse human capabilities is significant for its practical applications. This work proposes an enhanced LfI approach for learning the optimal RL policy by leveraging multimodal human behaviors in the setting of N-driver concurrent interventions. Specifically, we first learn the N number of human digital drivers from the multi-human demonstration dataset, wherein each driver possesses its own policy distribution. Then, the post-trained drivers will be kept in the training loop of the RL algorithms, providing diverse driving guidance whenever the intervention is required. Additionally, to better utilize the provided guidance, we augment the RL regarding the fundamental architecture and optimization objectives to facilitate the proposed uncertainty-aware reinforcement learning (UnaRL) algorithm. The proposed approach, which won 2nd place in the Alibaba Future Car Innovation Challenge 2022, is solidly compared in two challenging autonomous driving scenarios against state-of-the-art (SOTA) LfI baselines, and results of both simulation and real-world experiment confirm the superiority of our method in terms of learning robustness and driving performance. Videos and source code are provided.1

3D object-level mapping is a fundamental problem in robotics, which is especially challenging when object CAD models are unavailable during inference. We propose a framework that can reconstruct high-quality object-level maps for unknown objects. Our approach takes multiple RGB-D images as input and outputs dense 3D shapes and 9-DoF poses (including 3 scale parameters) for detected objects. The core idea is to leverage a learnt generative model for a category of object shapes as priors and to formulate a probabilistic, uncertainty-aware optimization framework for 3D reconstruction. We derive a probabilistic formulation that propagates shape and pose uncertainty through two novel loss functions. Unlike current state-of-the-art approaches, we explicitly model the uncertainty of the object shapes and poses during our optimization, resulting in a high-quality object-level mapping system. Moreover, the estimated shape and pose uncertainties, which we demonstrate can accurately reflect the true errors of our object maps, can be useful for downstream robotics tasks such as active vision. We perform extensive evaluations on indoor and outdoor real-world datasets, achieving substantial improvements over state-of-the-art methods. Our code is available at https://github.com/TRAILab/UncertainShapePose.

Continuum Dexterous Manipulators (CDMs) are well-suited tools for minimally invasive surgery due to their inherent dexterity and reachability. Nonetheless, their flexible structure and non-linear curvature pose significant challenges for shape-based feedback control. The use of Fiber Bragg Grating (FBG) sensors for shape sensing has shown great potential in estimating the CDM’s tip position and subsequently reconstructing the shape using optimization algorithms. This optimization, however, is under-constrained and may be ill-posed for complex shapes, falling into local minima. In this work, we introduce a novel method capable of directly estimating a CDM’s shape from FBG sensor wavelengths using a deep neural network. In addition, we propose the integration of uncertainty estimation to address the critical issue of uncertainty in neural network predictions. Neural network predictions are unreliable when the input sample is outside the training distribution or corrupted by noise. Recognizing such deviations is crucial when integrating neural networks within surgical robotics, as inaccurate estimations can pose serious risks to the patient. We present a robust method that not only improves the precision upon existing techniques for FBG-based shape estimation but also incorporates a mechanism to quantify the models’ confidence through uncertainty estimation. We validate the uncertainty estimation through extensive experiments, demonstrating its effectiveness and reliability on out-of-distribution (OOD) data, adding an additional layer of safety and precision to minimally invasive surgical robotics.

The injection of therapeutic agents into the sub-retinal space might allow improved treatment of age-related macular degeneration. Various robotic systems have been developed to achieve the required precision and, in combination with intraoperative Optical Coherence Tomography (iOCT) imaging, methods for autonomous robotic guidance have been proposed. In such systems, the robot’s cognition is often governed by machine learning algorithms, such as convolutional neural networks (CNNs), which provide semantic scene information from iOCT images. Although the robot performs a surgical task autonomously, human supervision is critical to monitor the robot’s execution and, if necessary, stop the robot or take control to avoid trauma to the patient. In this paper, we propose a novel visualization concept for improved human supervision of autonomous robotic subretinal injection that integrates uncertainty information of the data provided to the robot. We design a focus and context visualization that renders an automatically identified instrument-aligned B-scan in the context of the 3D OCT volume. Our visualization is enriched by augmenting the uncertainty information on the instrument-aligned B-scan. To dynamically model task-specific uncertainty, we introduce a weighting scheme to assign an importance factor to each pair of classes, controlling the impact of their confusion on the overall uncertainty. We demonstrate our visualization concept on iOCT volumes acquired at different stages during subretinal injection on ex-vivo porcine eyes. We show that our processing pipeline achieves sufficient update rates for surgical display and discuss the impact of our visualization concept on the acceptance of robotic task autonomy for subretinal injection procedures.

In-flight high-speed object capturing is crucial in nature to improve survival and adaptation to the environment, such as the predation of frogs, leopards, and eagles. Despite its ubiquitousness in nature, capturing fast-moving objects is extremely challenging in engineering implementations. In this paper, we report an ultrafast gripper based on tunable bistable structures. Different from current designs which are only suitable for objects with certain speed ranges once the grippers are fabricated, the working range of object speed of the proposed gripper could be reprogrammed by controlling the sensitivity of the structures. We present the design and fabrication of the proposed gripper in detail. A theoretical model is introduced to construct the energy landscape of the structures and the force response of the gripper when programmed to different states. The results show that in the original state, the gripper is capable of capturing a flying table tennis ball with a high speed of 15 m/s in only 6 ms. When the proposed gripper is controlled to the ultra-sensitive state, a flying ball with only 1 m/s could also be captured. This work broadens the frontiers of in-flight capturing design, and we envision broader promising applications.

In this paper, we strongly advocate square-root covariance (instead of information) filtering for Visual-Inertial Navigation Systems (VINS), in particular on resource-constrained edge devices, because of its superior efficiency and numerical stability. Although VINS have made tremendous progress in recent years, they still face resource stringency and numerical instability on embedded systems when imposing limited word length. To overcome these challenges, we develop an ultrafast and numerically-stable square-root filter (SRF)-based VINS algorithm (i.e., SR-VINS). The numerical stability of the proposed SR-VINS is inherited from the adoption of square-root covariance while the remarkable efficiency is largely enabled by the novel SRF update method that is based on our new permuted-QR (P-QR), which fully utilizes and properly maintains the upper triangular structure of the square-root covariance matrix. Furthermore, we choose a special ordering of the state variables which is amenable for (P-)QR operations in the SRF propagation and update and prevents unnecessary computation. The proposed SR-VINS is validated extensively through numerical studies, demonstrating that when the state-of-the-art (SOTA) filters have numerical difficulties, our SR-VINS has superior numerical stability, and remarkably, achieves efficient and robust performance on 32-bit single-precision float at a speed nearly twice as fast as the SOTA methods. We also conduct comprehensive real-world experiments to validate the efficiency, accuracy, and robustness of the proposed SR-VINS.

Traversability estimation in rugged, unstructured environments remains a challenging problem in field robotics. Often, the need for precise, accurate traversability estimation is in direct opposition to the limited sensing and compute capability present on affordable, small-scale mobile robots. To address this issue, we present a novel method to learn [u]ncertainty-aware [n]avigation features from high-fidelity scans of [real]-world environments (UNRealNet). This network can be deployed on-robot to predict these high-fidelity features using input from lower-quality sensors. UNRealNet predicts dense, metric-space features directly from single-frame lidar scans, thus reducing the effects of occlusion and odometry error. Our approach is label-free, and is able to produce traversability estimates that are robot-agnostic. Additionally, we can leverage UNRealNet’s predictive uncertainty to both produce risk-aware traversability estimates, and refine our feature predictions over time. We find that our method outperforms traditional local mapping and inpainting baselines by up to 40%, and demonstrate its efficacy on multiple legged platforms.

This paper proposes a novel sensor fusion based on Unscented Kalman Filtering for the online estimation of joint-torques of humanoid robots without joint-torque sensors. At the feature level, the proposed approach considers multimodal measurements (e.g. currents, accelerations, etc.) and non-directly measurable effects, such as external contacts, thus leading to joint torques readily usable in control architectures for human-robot interaction. The proposed sensor fusion can also integrate distributed, non-collocated force/torque sensors, thus being a flexible framework with respect to the underlying robot sensor suit. To validate the approach, we show how the proposed sensor fusion can be integrated into a two-level torque control architecture aiming at task-space torque-control. The performances of the proposed approach are shown through extensive tests on the new humanoid robot ergoCub, currently being developed at Istituto Italiano di Tecnologia. We also compare our strategy with the existing state-of-the-art approach based on the recursive Newton-Euler algorithm. Results demonstrate that our method achieves low root mean square errors in torque tracking, ranging from 0.05 Nm to 2.5 Nm, even in the presence of external contacts.

Autonomous navigation in the underwater environment is challenging due to limited visibility, dynamic changes, and the lack of a cost-efficient, accurate localization system. We introduce UIVNAV, a novel end-to-end underwater navigation solution designed to navigate robots over Objects of Interest (OOI) while avoiding obstacles, all without relying on localization. UIVNAV utilizes imitation learning and draws inspiration from the navigation strategies employed by human divers, who do not rely on localization. UIVNAV consists of the following phases: (1) generating an intermediate representation (IR) and (2) training the navigation policy based on human-labeled IR. By training the navigation policy on IR instead of raw data, the second phase is domain-invariant — the navigation policy does not need to be retrained if the domain or the OOI changes. We demonstrate this within simulation by deploying the same navigation policy to survey two distinct Objects of Interest (OOIs): oyster and rock reefs. We compared our method with complete coverage and random walk methods, showing that our approach is more efficient in gathering information for OOIs while avoiding obstacles. The results show that UIVNAV chooses to visit the areas with larger area sizes of oysters or rocks with no prior information about the environment or localization. Moreover, a robot using UIVNAV compared to complete coverage method surveys on average 36% more oysters when traveling the same distances. We also demonstrate the feasibility of real-time deployment of UIVNAV in pool experiments with BlueROV underwater robot for surveying a bed of oyster shells.

Tremendous variations coupled with large degrees of freedom in UAV-based imaging conditions lead to a significant lack of data in adequately learning UAV-based perception models. Using various synthetic renderers in conjunction with perception models is prevalent to create synthetic data to augment the learning in the ground-based imaging domain. However, severe challenges in the austere UAV-based domain require distinctive solutions to image synthesis for data augmentation. In this work, we leverage recent advancements in neural rendering to improve static and dynamic novel-view UAV-based image synthesis, especially from high altitudes, capturing salient scene attributes. Finally, we demonstrate a considerable performance boost is achieved when a state-of-the-art detection model is optimized primarily on hybrid sets of real and synthetic data instead of the real or synthetic data separately.

Controlled execution of dynamic motions in quadrupedal robots, especially those with articulated soft bodies, presents a unique set of challenges that traditional methods struggle to address efficiently. In this study, we tackle these issues by relying on a simple yet effective two-stage learning framework to generate dynamic motions for quadrupedal robots. First, a gradient-free evolution strategy is employed to discover simply represented control policies, eliminating the need for a predefined reference motion. Then, we refine these policies using deep reinforcement learning. Our approach enables the acquisition of complex motions like pronking and back-flipping, effectively from scratch. Additionally, our method simplifies the traditionally labour-intensive task of reward shaping, boosting the efficiency of the learning process. Importantly, our framework proves particularly effective for articulated soft quadrupeds, whose inherent compliance and adaptability make them ideal for dynamic tasks but also introduce unique control challenges.

Accurate disturbance estimation is essential for safe robot operations. The recently proposed neural moving horizon estimation (NeuroMHE), which uses a portable neural network to model the MHE’s weightings, has shown promise in further pushing the accuracy and efficiency boundary. Currently, NeuroMHE is trained through gradient descent, with its gradient computed recursively using a Kalman filter. This paper proposes a trust-region policy optimization method for training NeuroMHE. We achieve this by providing the second-order derivatives of MHE, referred to as the MHE Hessian. Remarkably, we show that many of the intermediate results used to obtain the gradient, especially the Kalman filter, can be efficiently reused to compute the MHE Hessian. This offers linear computational complexity with respect to the MHE horizon. As a case study, we evaluate the proposed trust region NeuroMHE on real quadrotor flight data for disturbance estimation. Our approach demonstrates highly efficient training in under 5 min using only 100 data points. It outperforms a state-of-the-art neural estimator by up to 68.1% in force estimation accuracy, utilizing only 1.4% of its network parameters. Furthermore, our method showcases enhanced robustness to network initialization compared to the gradient descent counterpart.

Recent work has considered trust-aware decision making for human-robot collaboration (HRC) with a focus on model learning. In this paper, we are interested in enabling the HRC system to complete complex tasks specified using temporal logic formulas that involve human trust. Since accurately observing human trust in robots is challenging, we adopt the widely used partially observable Markov decision process (POMDP) framework for modelling the interactions between humans and robots. To specify the desired behaviour, we propose to use syntactically co-safe linear distribution temporal logic (scLDTL), a logic that is defined over predicates of states as well as belief states of partially observable systems. The incorporation of belief predicates in scLDTL enhances its expressiveness while simultaneously introducing added complexity. This also presents a new challenge as the belief predicates must be evaluated over the continuous (infinite) belief space. To address this challenge, we present an algorithm for solving the optimal policy synthesis problem. First, we enhance the belief MDP (derived by reformulating the POMDP) with a probabilistic labelling function. Then a product belief MDP is constructed between the probabilistically labelled belief MDP and the automaton translation of the scLDTL formula. Finally, we show that the optimal policy can be obtained by leveraging existing point-based value iteration algorithms with essential modifications. Human subject experiments with 21 participants on a driving simulator demonstrate the effectiveness of the proposed approach.

Collaboration between humans and robots is becoming increasingly crucial in our daily life. In order to accomplish efficient cooperation, trust recognition is vital, empowering robots to predict human behaviors and make trust-aware decisions. Consequently, there is an urgent need for a generalized approach to recognize human-robot trust. This study addresses this need by introducing an EEG-based method for trust recognition during human-robot cooperation. A human-robot cooperation game scenario is used to stimulate various human trust levels when working with robots. To enhance recognition performance, the study proposes an EEG Vision Transformer model coupled with a 3-D spatial representation to capture the spatial information of EEG, taking into account the topological relationship among electrodes. To validate this approach, a public EEG-based human trust dataset called EEGTrust is constructed. Experimental results indicate the effectiveness of the proposed approach, achieving an accuracy of 74.99% in slice-wise cross-validation and 62.00% in trial-wise cross-validation. This outperforms baseline models in both recognition accuracy and generalization. Furthermore, an ablation study demonstrates a significant improvement in trust recognition performance of the spatial representation. The source code and EEGTrust dataset are available at https://github.com/CaiyueXu/EEGTrust.

Data collection for forestry, timber, and agriculture relies on manual techniques which are labor-intensive and time-consuming. We seek to demonstrate that robotics offers improvements over these techniques and can accelerate agricultural research, beginning with semantic segmentation and diameter estimation of trees in forests and orchards. We present TreeScope v1.0, the first robotics dataset for precision agriculture and forestry addressing the counting and mapping of trees in forestry and orchards. TreeScope provides LiDAR data from agricultural environments collected with robotics platforms, such as UAV and mobile robot platforms carried by vehicles and human operators. In the first release of this dataset, we provide ground-truth data with over 1,800 manually annotated semantic labels for tree stems and field-measured tree diameters. We share benchmark scripts for these tasks that researchers may use to evaluate the accuracy of their algorithms. Finally, we run our open-source diameter estimation and off-the-shelf semantic segmentation algorithms and share our baseline results.The dataset can be found at https://treescope.org, and the data pre-processing and benchmark code is available at https://github.com/KumarRobotics/treescope.

A novel algorithm to solve the 2D k-shortest non-homotopic path planning (k-SNPP) task is proposed in this paper. The task is of practical significance as a sub-module for higherlevel planning and scheduling tasks, and is gaining increasing attention and focus in recent years. There have existed algorithms that explicitly characterised non-homotopic paths using topological invariants such as ℎ-signature and winding number. However, these algorithms are inefficient due to their separate treatment of topology and geometry: Topological invariants are singularly utilised for distinguishing non-homotopic property among paths, which significantly increases the volume of the robot configuration space. Meanwhile, distance-optimal path planners search for locally shortest paths in the augmented space, which becomes extremely time-consuming. In this paper, a topological tree is proposed to simultaneously leverage topology and geometry. The tree grows from the starting location and explores all topological routes, until the best k of its leaves reach the goal. It is proven that different branches of the tree explore different homotopy classes of paths, and all the branches are locally shortest. Comparative experiments for k-SNPP are conducted in challenging grid-based simulated environments to validate the performance of the proposed algorithm. The C++ implementation of the proposed algorithm is released for the benefit of the robotics community.