ICRA 2024 Papers

Accurate understanding of lunar surface topography is vital for effective decision-making and remote control of lunar rovers during exploration missions. Conventional sensing methods often struggle to capture the intricate details of the lunar landscape. In response, we propose an innovative approach that leverages NeRF to synthesize new viewpoints within the expansive lunar environment. By blending 3D hash grids and 2D plane grids representations, our approach provides a comprehensive scene representation. We employ the technique of spiral sampling and feature rendering to enhance rendering quality while simultaneously reducing training time. Additionally, we leverage sparse point cloud to aid the model in better learning the geometric structure of the lunar environment. Through experimentation, we have demonstrated that our method is capable of synthesizing realistic images of lunar environments.

Model predictive control (MPC) may provide local motion planning for mobile robotic platforms. The challenging aspect is the analytic representation of collision cost for the case when both the obstacle map and robot footprint are arbitrary. We propose a Neural Potential Field: a neural network model that returns a differentiable collision cost based on robot pose, obstacle map, and robot footprint. The differentiability of our model allows its usage within the MPC solver. It is computationally hard to solve problems with a very high number of parameters. Therefore, our architecture includes neural image encoders, which transform obstacle maps and robot footprints into embeddings, which reduce problem dimensionality by two orders of magnitude. The reference data for network training are generated based on algorithmic calculation of a signed distance function. Comparative experiments showed that the proposed approach is comparable with existing local planners: it provides trajectories with outperforming smoothness, comparable path length, and safe distance from obstacles.

Sampling-based planning algorithms like Rapidly-exploring Random Tree (RRT) are versatile in solving path planning problems. RRT* offers asymptotic optimality but requires growing the tree uniformly over the free space, which leaves room for efficiency improvement. To accelerate convergence, rule-based informed approaches sample states in an admissible ellipsoidal subset of the space determined by the current path cost. Learning-based alternatives model the topology of the free space and infer the states close to the optimal path to guide planning. We propose Neural Informed RRT* to combine the strengths from both sides. We define point cloud representations of free states. We perform Neural Focus, which constrains the point cloud within the admissible ellipsoidal subset from Informed RRT*, and feeds into PointNet++ for refined guidance state inference. In addition, we introduce Neural Connect to build connectivity of the guidance state set and further boost performance in challenging planning problems. Our method surpasses previous works in path planning benchmarks while preserving probabilistic completeness and asymptotic optimality. We deploy our method on a mobile robot and demonstrate real world navigation around static obstacles and dynamic humans. Code is available at https://github.com/tedhuang96/nirrt_star.

Collision detection is one of the most time-consuming operations during motion planning. Thus, there is an increasing interest in exploring machine learning techniques to speed up collision detection and sampling-based motion planning. A recent line of research focuses on utilizing neural signed distance functions of either the robot geometry or the swept volume of the robot motion. Building on this, we present a novel neural implicit swept volume model to continuously represent arbitrary motions parameterized by their start and goal configurations. This allows to quickly compute signed distances for any point in the task space to the robot motion. Further, we present an algorithm combining the speed of the deep learning-based signed distance computations with the strong accuracy guarantees of geometric collision checkers. We validate our approach in simulated and real-world robotic experiments, and demonstrate that it is able to speed up a commercial bin picking application.

Achieving efficient and consistent localization with a prior map remains challenging in robotics. Conventional keyframe-based approaches often suffer from sub-optimal viewpoints due to limited field of view (FOV) and/or constrained motion, thus degrading the localization performance. To address this issue, we design a real-time tightly-coupled Neural Radiance Fields (NeRF)-aided visual-inertial navigation system (VINS). In particular, by effectively leveraging the NeRF’s potential to synthesize novel views, the proposed NeRF-VINS overcomes the limitations of traditional keyframe-based maps (with limited views) and optimally fuses IMU, monocular images, and synthetically rendered images within an efficient filter-based framework. This tightly-coupled fusion enables efficient 3D motion tracking with bounded errors. We extensively validate the proposed NeRF-VINS against the state-of-the-art methods that use prior map information, and demonstrate its ability to perform real-time localization, at 15 Hz, on a resource-constrained Jetson AGX Orin embedded platform.

In various applications, such as robotic navigation and remote visual assistance, expanding the field of view (FOV) of the camera proves beneficial for enhancing environmental perception. Unlike image outpainting techniques aimed solely at generating aesthetically pleasing visuals, these applications demand an extended view that faithfully represents the scene. To achieve this, we formulate a new problem of faithful FOV extrapolation that utilizes a set of pre-captured images as prior knowledge of the scene. To address this problem, we present a simple yet effective solution called NeRF-Enhanced Outpainting (NEO) that uses extended-FOV images generated through NeRF to train a scene-specific image outpainting model. To assess the performance of NEO, we conduct comprehensive evaluations on three photorealistic datasets and one real-world dataset. Extensive experiments on the benchmark datasets showcase the robustness and potential of our method in addressing this challenge. We believe our work lays a strong foundation for future exploration within the research community.

Reinforcement learning has shown great potential in improving the performance of robot navigation. In response to the increasing deployments of mobile robots within various scenarios, a data-driven paradigm of navigation approach with safety verification is preferred where one can train RL algorithms with large amounts of prior data, keep learning continuously, and ensure safe navigation in applications. Conventional end-to-end reinforcement learning navigation paradigms have encountered multiple challenges in meeting these demands. In this work, we introduce a novel robot navigation approach termed NaviFormer. This approach handles navigation tasks based on sequence modeling to obtain the data-driven ability. It also integrates rule-based verification for safety insurance. We conduct a series of experiments to validate the data-driven ability of our approach and to compare it with existing navigation methods. We also perform quantitative tests on a real-world robot platform, TurtleBot. The experimental results show our method’s outstanding data-driven ability and highlight its superior arrival rate and generalization compared to other state-of-the-art methods like the PPO-based navigation method.

Wirelessly actuated magnetic microrobots are promising tools in medical applications due to their tiny sizes and attractive robotic properties. However, it remains a huge challenge to integrate sufficient functionalities in a limited volume. Microscopic natural phenomenon is a great reference for current microrobot design, where the underlying intelligence and subtlety spurs related modern artificial systems. Inspired by air bubbles in nature, herein, we report a kind of novel magnetic air bubble microrobots. The air bubble-based structure enables multiple functionalities including cargo delivery, multimode locomotion, micromanipulation, medical imaging, and biosensing. The proposed microrobot is essentially Pickering bubbles composed of magnetic particles and air bubbles. Their hollow structures help produce lighter microrobots with density less than 1 g/cm3, enabling buoyancy-based self-propulsion. Buoyancy and magnetic forces actuation enables flexible 3D locomotion in fluidic environments. Experimental results show that the microrobots can be controlled properly for designated assignments. Furthermore, the introduction of air bubble enhances ultrasound imaging, facilitating further in vivo applications. These findings offer a significant microrobot design paradigm by exploiting natural physical intelligence at the small scale.

Visual Place Recognition (VPR) in indoor environments is beneficial to humans and robots for better localization and navigation. It is challenging due to appearance changes at various frequencies, and difficulties of obtaining ground truth metric trajectories for training and evaluation. This paper introduces the NYC-Indoor-VPR dataset, a unique and rich collection of over 36,000 images compiled from 13 distinct crowded scenes in New York City taken under varying lighting conditions with appearance changes. Each scene has multiple revisits across a year. To establish the ground truth for VPR, we propose a semiautomatic annotation approach that computes the positional information of each image. Our method specifically takes pairs of videos as input and yields matched pairs of images along with their estimated relative locations. The accuracy of this matching is refined by human annotators, who utilize our annotation software to correlate the selected keyframes. Finally, we present a benchmark evaluation of several state-of-the-art VPR algorithms using our annotated dataset, revealing its challenge and thus value for VPR research.

Fuel efficiency is a crucial aspect of long-distance cargo transportation by oil-powered trucks that economize on costs and decrease carbon emissions. Current predictive control methods depend on an accurate model of vehicle dynamics and engine, including weight, drag coefficient, and the Brake-specific Fuel Consumption (BSFC) map of the engine. We propose a pure data-driven method, Neural Predictive Control (NPC), which does not use any physical model for the vehicle. After training with over 20,000 km of historical data, the novel proposed NVFormer implicitly models the relationship between vehicle dynamics, road slope, fuel consumption, and control commands using the attention mechanism. Based on the online sampled primitives from the past of the current freight trip and anchor-based future data synthesis, the NVFormer can infer optimal control command for reasonable fuel consumption. The physical model-free NPC outperforms the base PCC method with 2.41% and 3.45% more significant fuel saving in simulation and open-road highway testing, respectively.

In this work, we investigate task planning for mobile robots under linear temporal logic (LTL) specifications. This problem is particularly challenging when robots navigate in continuous workspaces due to the high computational complexity involved. Sampling-based methods have emerged as a promising avenue for addressing this challenge by incrementally constructing random trees, thereby sidestepping the need to explicitly explore the entire state-space. However, the performance of this sampling-based approach hinges crucially on the chosen sampling strategy, and a well-informed heuristic can notably enhance sample efficiency. In this work, we propose a novel neural-network guided (NN-guided) sampling strategy tailored for LTL planning. Specifically, we employ a multi-modal neural network capable of extracting features concurrently from both the workspace and the Büchi automaton. This neural network generates predictions that serve as guidance for random tree construction, directing the sampling process toward more optimal directions. Through numerical experiments, we compare our approach with existing methods and demonstrate its superior efficiency, requiring less than 15% of the time of the existing methods to find a feasible solution.

Neural implicit representations have emerged as a promising solution for providing dense geometry in Simultaneous Localization and Mapping (SLAM). However, existing methods in this direction fall short in terms of global consistency and low latency. This paper presents NGEL-SLAM to tackle the above challenges. To ensure global consistency, our system leverages a traditional feature-based tracking module that incorporates loop closure. Additionally, we maintain a global consistent map by representing the scene using multiple neural implicit fields, enabling quick adjustment to the loop closure. Moreover, our system allows for fast convergence through the use of octree-based implicit representations. The combination of rapid response to loop closure and fast convergence makes our system a truly low-latency system that achieves global consistency. Our system enables rendering high-fidelity RGB-D images, along with extracting dense and complete surfaces. Experiments on both synthetic and real-world datasets suggest that our system achieves state-of-the-art tracking and mapping accuracy while maintaining low latency.

Image correspondence serves as the backbone for many tasks in robotics, such as visual fusion, localization, and mapping. However, existing correspondence methods do not scale to large multi-robot systems, and they struggle when image features are weak, ambiguous, or evolving. In response, we propose Natural Quick Response codes, or N-QR, which enables rapid and reliable correspondence between large-scale teams of heterogeneous robots. Our method works like a QR code, using keypoint-based alignment, rapid encoding, and error correction via ensembles of image patches of natural patterns. We deploy our algorithm in a production-scale robotic farm, where groups of growing plants must be matched across many robots. We demonstrate superior performance compared to several baselines, obtaining a retrieval accuracy of 88.2%. Our method generalizes to a farm with 100 robots, achieving a 12.5x reduction in bandwidth and a 20.5x speedup. We leverage our method to correspond 700k plants and confirm a link between a robotic seeding policy and germination.

This paper introduces a Nonlinear Model Predictive Control (N-MPC) framework exploiting a Deep Neural Network for processing onboard-captured depth images for collision avoidance in trajectory-tracking tasks with UAVs. The network is trained on simulated depth images to output a collision score for queried 3D points within the sensor field of view. Then, this network is translated into an algebraic symbolic equation and included in the N-MPC, explicitly constraining predicted positions to be collision-free throughout the receding horizon. The N-MPC achieves real time control of a UAV with a control frequency of 100Hz. The proposed framework is validated through statistical analysis of the collision classifier network, as well as Gazebo simulations and real experiments to assess the resulting capabilities of the N-MPC to effectively avoid collisions in cluttered environments. The associated code is released open-source.

In the expanding landscape of AI-enabled robotics, robust quantification of predictive uncertainties is of great importance. Three-dimensional (3D) object detection, a critical robotics operation, has seen significant advancements; however, the majority of current works focus only on accuracy and ignore uncertainty quantification. Addressing this gap, our novel study integrates the principles of conformal inference (CI) with information theoretic measures to perform lightweight, Monte Carlo-free uncertainty estimation within a multimodal framework. Through a multivariate Gaussian product of the latent variables in a Variational Autoencoder (VAE), features from RGB camera and LiDAR sensor data are fused to improve the prediction accuracy. Normalized mutual information (NMI) is leveraged as a modulator for calibrating uncertainty bounds derived from CI based on a weighted loss function. Our simulation results show an inverse correlation between inherent predictive uncertainty and NMI throughout the model’s training. The framework demonstrates comparable or better performance in KITTI 3D object detection benchmarks to similar methods that are not uncertainty-aware, making it suitable for real-time edge robotics.

This paper presents an efficient method for updating particles in a particle filter (PF) to address the position estimation problem when dealing with sharp-peaked likelihood functions derived from multiple observations. Sharp-peaked likelihood functions commonly arise from millimeter-accurate distance observations of carrier phases in the global navigation satellite system (GNSS). However, when such likelihood functions are used for particle weight updates, the absence of particles within the peaks leads to all particle weights becoming zero. To overcome this problem, in this study, a straightforward and effective approach is introduced for updating particles when dealing with sharp-peaked likelihood functions obtained from multiple observations. The proposed method, termed as the multiple update PF, leverages prior knowledge regarding the spread of distribution for each likelihood function and conducts weight updates and resampling iteratively in the particle update process, prioritizing the likelihood function spreads. Experimental results demonstrate the efficacy of our proposed method, particularly when applied to position estimation utilizing GNSS pseudorange and carrier phase observations. The multiple update PF exhibits faster convergence with fewer particles when compared to the conventional PF. Moreover, vehicle position estimation experiments conducted in urban environments reveal that the proposed method outperforms conventional GNSS positioning techniques, yielding more accurate position estimates.

The rapidly evolving field of robotics necessitates methods that can facilitate the fusion of multiple modalities. Specifically, when it comes to interacting with tangible objects, effectively combining visual and tactile sensory data is key to understanding and navigating the complex dynamics of the physical world, enabling a more nuanced and adaptable response to changing environments. Nevertheless, much of the earlier work in merging these two sensory modalities has relied on supervised methods utilizing datasets labeled by humans. This paper introduces MViTac, a novel methodology that leverages contrastive learning to integrate vision and touch sensations in a self-supervised fashion. By availing both sensory inputs, MViTac leverages intra and inter-modality losses for learning representations, resulting in enhanced material property classification and more adept grasping prediction. Through a series of experiments, we showcase the effectiveness of our method and its superiority over existing state-of-the-art self-supervised and supervised techniques. In evaluating our methodology, we focus on two distinct tasks: material classification and grasping success prediction. Our results indicate that MViTac facilitates the development of improved modality encoders, yielding more robust representations as evidenced by linear probing assessments. https://sites.google.com/view/mvitac/home

Real-time recognition and prediction of surgical activities are fundamental to advancing safety and autonomy in robot-assisted surgery. This paper presents a multimodal transformer architecture for real-time recognition and prediction of surgical gestures and trajectories based on short segments of kinematic and video data. We conduct an ablation study to evaluate the impact of fusing different input modalities and their representations on gesture recognition and prediction performance. We perform an end-to-end assessment of the proposed architecture using the JHU-ISI Gesture and Skill Assessment Working Set (JIGSAWS) dataset. Our model outperforms the state-of-the-art (SOTA) with 89.5% accuracy for gesture prediction through effective fusion of kinematic features with spatial and contextual video features. It achieves the real-time performance of 1.1-1.3ms for processing a 1-second input window by relying on a computationally efficient model.

3D object detection models that exploit both LiDAR and camera sensor features are top performers in large-scale autonomous driving benchmarks. A transformer is a popular network architecture used for this task, in which so-called object queries act as candidate objects. Initializing these object queries based on current sensor inputs is a common practice. For this, existing methods strongly rely on LiDAR data however, and do not fully exploit image features. Besides, they introduce significant latency. To overcome these limitations we propose EfficientQ3M, an efficient, modular, and multimodal solution for object query initialization for transformer-based 3D object detection models. The proposed initialization method is combined with a "modality-balanced" transformer decoder where the queries can access all sensor modalities throughout the decoder. In experiments, we outperform the state of the art in transformer-based LiDAR object detection on the competitive nuScenes benchmark and showcase the benefits of input-dependent multimodal query initialization, while being more efficient than the available alternatives for LiDAR-camera initialization. The proposed method can be applied with any combination of sensor modalities as input, demonstrating its modularity.