ICRA 2024 论文列表

The effective reasoning of integrated multimodal perception information is crucial for achieving enhanced end-to-end autonomous driving performance. In this paper, we introduce a novel multitask imitation learning framework for end-to-end autonomous driving that leverages a dual attention transformer (DualAT) to enhance the multimodal fusion and waypoint prediction processes. A self-attention mechanism captures global context information and models the long-term temporal dependencies of waypoints for multiple time steps. On the other hand, a cross-attention mechanism implicitly associates the latent feature representations derived from different modalities through a learnable geometrically linked positional embedding. Specifically, the DualAT excels at processing and fusing information from multiple camera views and LiDAR sensors, enabling comprehensive scene understanding for multitask learning. Furthermore, the DualAT introduces a novel waypoint prediction architecture that combines the temporal relationships between waypoints with the spatial features extracted from sensor inputs. We evaluate our approach on both the Town05 and Longest6 benchmarks using the closed-loop CARLA urban driving simulator and provide extensive ablation studies. The experimental results demonstrate that our approach significantly outperforms the state-of-the-art methods.

To foster an immersive and natural human-robot interaction (HRI), the implementation of tactile perception and feedback becomes imperative, effectively bridging the conventional sensory gap. In this paper, we propose a dual-modal electronic skin (e-skin) that integrates magnetic tactile sensing and vibration feedback for enhanced HRI. The dual-modal tactile e-skin offers multi-functional tactile sensing and programmable haptic feedback, underpinned by a layered structure comprised of flexible magnetic films, soft silicone elastomer, a Hall sensor and actuator array, and a microcontroller unit. The e-skin captures the magnetic field changes caused by subtle deformations through Hall sensors, employing deep learning for accurate tactile perception. Simultaneously, the actuator array generates mechanical vibrations to facilitate haptic feedback, delivering diverse mechanical stimuli. Notably, the dual-modal e-skin is capable of transmitting tactile information bidirectionally, enabling object recognition and fine-weighing operations. This bidirectional tactile interaction framework will enhance the immersion and efficiency of interactions between humans and robots.

In this paper, we address the problem of relative localization of two mobile agents. Specifically, we consider the Dual-IMU system, where each agent is equipped with one IMU, and employs relative pose observations between them. Previous works, however, typically assumed known ego motion and ignored biases of the IMUs. Instead, we study the most general case of unknown biases for both IMUs. Besides the derivation of dynamic model equations of the proposed system, we focus on the observability analysis, for the observability under general motion and the unobservable directions arising from various special motions. Through numerical simulations, we validate our key observability findings and examine their impact on the estimation accuracy and consistency. Finally, the system is implemented to achieve effective relative localization of an HMD with respect to a vehicle moving in the real world.

Robotic push-grasping in densely cluttered environments presents significant challenges due to unbalanced synergy and redundancy between both actions, leading to decreased grasp efficiency. In this paper, a novel double-critic deep reinforcement learning framework is introduced to optimize the push-grasping synergy for robotic manipulation in such environments, aiming to significantly reduce pre-grasping redundancy. This framework incorporates two distinct Deep Q-learning critics: Critic I selects the best course of actions based on the current state derived from visual interpretation, whereas Critic II evaluates the success rate of the current state-action pairing. To further refine the push-grasping synergy, an active double-step learning mechanism is introduced to optimize the training reward function for the pushing action, thereby enhancing its effectiveness through increased intentionality. Simulations show that the proposed framework outperforms contemporary counterparts, notably in grasping success rate and action efficiency. Finally, the framework’s generalization and adaptability are demonstrated by conducting real-world experiments using novel objects without the need of retraining.

Modeling the kinematics and dynamics of robotics systems with suspended loads using dual quaternions has not been explored so far. This paper introduces a new innovative control strategy using dual quaternions for UAVs with cable-suspended loads, focusing on the sling load lifting and tracking problems. By utilizing the mathematical efficiency and compactness of dual quaternions, a unified representation of the UAV and its suspended load’s dynamics and kinematics is achieved, facilitating the realization of load lifting and trajectory tracking. The simulation results have tested the proposed strategy’s accuracy, efficiency, and robustness. This study makes a substantial contribution to present this novel control strategy that harnesses the benefits of dual quaternions for cargo UAVs. Our work also holds promise for inspiring future innovations in under-actuated systems control using dual quaternions.

Multi-object tracking (MOT) on static platforms, such as by surveillance cameras, has achieved significant progress, with various paradigms providing attractive performances. However, the effectiveness of traditional MOT methods is significantly reduced when it comes to dynamic platforms like drones. This decrease is attributed to the distinctive challenges in the MOT-on-drone scenario: (1) objects are generally small in the image plane, blurred, and frequently occluded, making them challenging to detect and recognize; (2) drones move and see objects from different angles, causing the unreliability of the predicted positions and feature embeddings of the objects. This paper proposes DroneMOT, which firstly proposes a Dual-domain Integrated Attention (DIA) module that considers the fast movements of drones to enhance the drone-based object detection and feature embedding for small-sized, blurred, and occluded objects. Then, an innovative Motion-Driven Association (MDA) scheme is introduced, considering the concurrent movements of both the drone and the objects. Within MDA, an Adaptive Feature Synchronization (AFS) technique is presented to update the object features seen from different angles. Additionally, a Dual Motion-based Prediction (DMP) method is employed to forecast the object positions. Finally, both the refined feature embeddings and the predicted positions are integrated to enhance the object association. Comprehensive evaluations on VisDrone2019-MOT and UAVDT datasets show that DroneMOT provides substantial performance improvements over the state-of-the-art in the domain of MOT on drones. The code will be available at https://github.com/PenK1nG/DroneMOT.

Large Language Models (LLMs) have shown promise in the autonomous driving sector, particularly in generalization and interpretability. We introduce a unique objectlevel multimodal LLM architecture that merges vectorized numeric modalities with a pre-trained LLM to improve context understanding in driving situations. We also present a new dataset of 160k QA pairs derived from 10k driving scenarios, paired with high quality control commands collected with RL agent and question answer pairs generated by teacher LLM (GPT-3.5). A distinct pretraining strategy is devised to align numeric vector modalities with static LLM representations using vector captioning language data. We also introduce an evaluation metric for Driving QA and demonstrate our LLM-driver’s proficiency in interpreting driving scenarios, answering questions, and decision-making. Our findings highlight the potential of LLM-based driving action generation in comparison to traditional behavioral cloning. We make our benchmark, datasets, and model available1 for further exploration.

As autonomous driving technology matures, end-to-end methodologies have emerged as a leading strategy, promising seamless integration from perception to control via deep learning. However, existing systems grapple with challenges such as unexpected open set environments and the complexity of black-box models. At the same time, the evolution of deep learning introduces larger, multimodal foundational models, offering multi-modal visual and textual understanding. In this paper, we harness these multimodal foundation models to enhance the robustness and adaptability of autonomous driving systems. We introduce a method to extract nuanced spatial features from transformers and the incorporation of latent space simulation for improved training and policy debugging. We use pixel/patch-aligned feature descriptors to expand foundational model capabilities to create an end-to-end multimodal driving model, demonstrating unparalleled results in diverse tests. Our solution combines language with visual perception and achieves significantly greater robustness on out-of-distribution situations.

We introduce Dream2Real, a robotics framework which integrates vision-language models (VLMs) trained on 2D data into a 3D object rearrangement pipeline. This is achieved by the robot autonomously constructing a 3D representation of the scene, where objects can be rearranged virtually and an image of the resulting arrangement rendered. These renders are evaluated by a VLM, so that the arrangement which best satisfies the user instruction is selected and recreated in the real world with pick-and-place. This enables language-conditioned rearrangement to be performed zero-shot, without needing to collect a training dataset of example arrangements. Results on a series of real-world tasks show that this framework is robust to distractors, controllable by language, capable of understanding complex multi-object relations, and readily applicable to both tabletop and 6-DoF rearrangement tasks. Videos are available on our webpage at: https://www.robot-learning.uk/dream2real.

We present a novel 3D odometry method that recovers the full motion of a vehicle only from a Doppler-capable range sensor. It leverages the radial velocities measured from the scene, estimating the sensor’s velocity from a single scan. The vehicle’s 3D motion, defined by its linear and angular velocities, is calculated taking into consideration its kinematic model which provides a constraint between the velocity measured at the sensor frame and the vehicle frame.Experiments carried out prove the viability of our single-sensor method compared to mounting an additional IMU. Our method provides a more reliable translation of the sensor, compared to the errors linked to IMUs due to noise and biases. Its short-term accuracy and fast operation (∼5ms) make it a proper candidate to supply the initialization to more complex localization algorithms or mapping pipelines. Not only does it reduce the error of the mapper, but it does so at a comparable level of accuracy as an IMU would. All without the need to mount and calibrate an extra sensor on the vehicle.

Robotic grasping refers to making a robotic system pick an object by applying forces and torques on its surface. Many recent studies use data-driven approaches to address grasping, but the sparse reward nature of this task made the learning process challenging to bootstrap. To avoid constraining the operational space, an increasing number of works propose grasping datasets to learn from. But most of them are limited to simulations. The present paper investigates how automatically generated grasps can be exploited in the real world. More than 7000 reach-and-grasp trajectories have been generated with Quality-Diversity (QD) methods on 3 different arms and grippers, including parallel fingers and a dexterous hand, and tested in the real world. Conducted analysis on the collected measure shows correlations between several Domain Randomization-based quality criteria and sim-to-real transferability. Key challenges regarding the reality gap for grasping have been identified, stressing matters on which researchers on grasping should focus in the future. A QD approach has finally been proposed for making grasps more robust to domain randomization, resulting in a transfer ratio of 84% on the Franka Research 3 arm.

Deploying machine learning algorithms for robot tasks in real-world applications presents a core challenge: overcoming the domain gap between the training and the deployment environment. This is particularly difficult for visuomotor policies that utilize high-dimensional images as input, particularly when those images are generated via simulation. A common method to tackle this issue is through domain randomization, which aims to broaden the span of the training distribution to cover the test-time distribution. However, this approach is only effective when the domain randomization encompasses the actual shifts in the test-time distribution. We take a different approach, where we make use of a single demonstration (a prompt) to learn policy that adapts to the testing target environment. Our proposed framework, PromptAdapt, leverages the Transformer architecture’s capacity to model sequential data to learn demonstration-conditioned visual policies, allowing for in-context adaptation to a target domain that is distinct from training. Our experiments in both simulation and real-world settings show that PromptAdapt is a strong domain-adapting policy that outperforms baseline methods by a large margin under a range of domain shifts, including variations in lighting, color, texture, and camera pose. Videos and more information can be viewed at project webpage: https://sites.google.com/view/promptadapt.

Dense visual correspondence plays a vital role in robotic perception. This work focuses on establishing the dense correspondence between a pair of images that captures dynamic scenes undergoing substantial transformations. We introduce Doduo to learn general dense visual correspondence from in-the-wild images and videos without ground truth supervision. Given a pair of images, it estimates the dense flow field encoding the displacement of each pixel in one image to its corresponding pixel in the other image. Doduo uses flow-based warping to acquire supervisory signals for the training. Incorporating semantic priors with self-supervised flow training, Doduo produces accurate dense correspondence robust to the dynamic changes of the scenes. Trained on an in-the-wild video dataset, Doduo illustrates superior performance on point-level correspondence estimation over existing self-supervised correspondence learning baselines. We also apply Doduo to articulation estimation and zero-shot goal-conditioned manipulation, underlining its practical applications in robotics. Code and additional visualizations are available at https://ut-austin-rpl.github.io/Doduo/

There is a current increase in the development of "4D" Doppler-capable radar and lidar range sensors that produce 3D point clouds where all points also have information about the radial velocity relative to the sensor. 4D radars in particular are interesting for object perception and navigation in low-visibility conditions (dust, smoke) where lidars and cameras typically fail. With the advent of high-resolution Doppler-capable radars comes the possibility of estimating odometry from single point clouds, foregoing the need for scan registration which is error-prone in feature-sparse field environments. We compare several odometry estimation methods, from direct integration of Doppler/IMU data and Kalman filter sensor fusion to 3D scan-to-scan and scan-to-map registration, on three datasets with data from two recent 4D radars and two IMUs. Surprisingly, our results show that the odometry from Doppler and IMU data alone give similar or better results than 3D point cloud registration. In our experiments, the position drift can be as low as 0.9% over 1.8 and 4.5km trajectories. That allows accurate estimation of 6-DOF ego-motion over long distances also in feature-sparse mine environments. These results are useful not least for applications of navigation with resource-constrained robot platforms in feature-sparse and low-visibility conditions such as mining, construction, and search & rescue operations.

With increasing numbers of mobile robots arriving in real-world applications, more robots coexist in the same space, interact, and possibly collaborate. Methods to provide such systems with system size scalability are known, for example, from swarm robotics. Example strategies are self-organizing behavior, a strict decentralized approach, and limiting the robot-robot communication. Despite applying such strategies, any multi-robot system breaks above a certain critical system size (i.e., number of robots) as too many robots share a resource (e.g., space, communication channel). We provide additional evidence based on simulations, that at these critical system sizes, the system performance separates into two phases: nearly optimal and minimal performance. We speculate that in real-world applications that are configured for optimal system size, the supposedly high-performing system may actually live on borrowed time as it is on a transient to breakdown. We provide two modeling options (based on queueing theory and a population model) that may help to support this reasoning.

When used in a real-world noisy environment, the capacity to generalize to multiple domains is essential for any autonomous scene text spotting system. However, existing state-of-the-art methods employ pretraining and fine-tuning strategies on natural scene datasets, which do not exploit the feature interaction across other complex domains. In this work, we explore and investigate the problem of domain-agnostic scene text spotting, i.e., training a model on multi-domain source data such that it can directly generalize to target domains rather than being specialized for a specific domain or scenario. In this regard, we present the community a text spotting validation benchmark called Under-Water Text (UWT) for noisy underwater scenes to establish an important case study. Moreover, we also design an efficient super-resolution based end-to-end transformer baseline called DA-TextSpotter which achieves comparable or superior performance over existing text spotting architectures for both regular and arbitrary-shaped scene text spotting benchmarks in terms of both accuracy and model efficiency. The dataset, code and pre-trained models have been released in our Github.

Accurate estimation of stereo camera extrinsic parameters is crucial to guarantee the performance of stereo matching algorithms. In prior arts, the online self-calibration of stereo cameras has commonly been formulated as a specialized visual odometry problem, without taking into account the principles of stereo rectification. In this paper, we first delve deeply into the concept of rectifying homography, which serves as the cornerstone for the development of our novel stereo camera online self-calibration algorithm, for cases where only a single pair of images is available. Furthermore, we introduce a simple yet effective solution for global optimum extrinsic parameter estimation in the presence of stereo video sequences. Additionally, we emphasize the impracticality of using three Euler angles and three components in the translation vectors for performance quantification. Instead, we introduce four new evaluation metrics to quantify the robustness and accuracy of extrinsic parameter estimation, applicable to both single-pair and multi-pair cases. Extensive experiments conducted across indoor and outdoor environments using various experimental setups validate the effectiveness of our proposed algorithm. The comprehensive evaluation results demonstrate its superior performance in comparison to the baseline algorithm. Our source code, demo video, and supplement are publicly available at mias.group/StereoCalibrator.

Safe corridor-based Trajectory Optimization (TO) presents an appealing approach for collision-free path planning of autonomous robots, because its convex formulation can guarantee global optimality. The safe corridor is constructed based on the obstacle map, however, the non-ideal perception induces uncertainty, which is rarely considered in the context of trajectory generation. In this paper, we propose Distributionally Robust Safe Corridor Constraints (DRSCCs) to consider the uncertainty of the safe corridor. Then, we integrate DRSCCs into the trajectory optimization framework using Bernstein basis polynomials. Theoretically, we rigorously prove that the proposed trajectory optimization problem is equivalent to a convex quadratic program, which is computationally efficient to deploy onto real robots. The simulation results show that our method enhances navigation safety by significantly reducing the infeasible motions compared to the baseline. Moreover, the proposed approach is validated through two robotic applications, a micro Unmanned Aerial Vehicle (UAV) and a quadruped robot Unitree A1.

Safety is a core challenge of autonomous robot motion planning, especially in the presence of dynamic and uncertain obstacles. Many recent results use learning and deep learning-based motion planners and prediction modules to predict multiple possible obstacle trajectories and generate obstacle-aware ego robot plans. However, planners that ignore the inherent uncertainties in such predictions incur collision risks and lack formal safety guarantees. In this paper, we present a computationally efficient safety filtering solution to reduce the collision risk of ego robot motion plans using multiple samples of obstacle trajectory predictions. The proposed approach reformulates the collision avoidance problem by computing safe halfspaces based on obstacle sample trajectories using distributionally robust optimization (DRO) techniques. The safe halfspaces are used in a model predictive control (MPC)-like safety filter to apply corrections to the reference ego trajectory thereby promoting safer planning. The efficacy and computational efficiency of our approach are demonstrated through numerical simulations.