ICRA 2023 论文列表

Manually specifying features that capture the diversity in traffic environments is impractical. Consequently, learning-based agents cannot realize their full potential as neural motion planners for autonomous vehicles. Instead, this work proposes to learn which features are task-relevant. Given its immediate relevance to motion planning, our proposed architecture encodes the probabilistic occupancy map as a proxy for obtaining pre-trained state representations of the environment. By leveraging a map-aware traffic graph formulation, our agent-centric encoder generalizes to arbitrary road networks and traffic situations. We show that our approach significantly improves the downstream performance of a reinforcement learning agent operating in urban traffic environments.

Miniaturized autonomous unmanned aerial vehicles (UAVs) are an emerging and trending topic. With their form factor as big as the palm of one hand, they can reach spots otherwise inaccessible to bigger robots and safely operate in human surroundings. The simple electronics aboard such robots (sub-100 mW) make them particularly cheap and attractive but pose significant challenges in enabling onboard sophisticated intelligence. In this work, we leverage a novel neural architecture search (NAS) technique to automatically identify several Pareto-optimal convolutional neural networks (CNNs) for a visual pose estimation task. Our work demonstrates how reallife and field-tested robotics applications can concretely leverage NAS technologies to automatically and efficiently optimize CNNs for the specific hardware constraints of small UAVs. We deploy several NAS-optimized CNNs and run them in closed-loop aboard a 27-g Crazyflie nano-UAV equipped with a parallel ultra-low power System-on-Chip. Our results improve the State-of-the-Art by reducing the in-field control error of 32% while achieving a real-time onboard inference-rate of ~10Hz@10mW and ~50Hz@90mW.

Correspondence identification (CoID) is an essential component for collaborative perception in multi-robot systems, such as connected autonomous vehicles. The goal of CoID is to identify the correspondence of objects observed by multiple robots in their own field of view in order for robots to consistently refer to the same objects. CoID is challenging due to perceptual aliasing, object non-covisibility, and noisy sensing. In this paper, we introduce a novel deep masked graph matching approach to enable CoID and address the challenges. Our approach formulates CoID as a graph matching problem and we design a masked neural network to integrate the multimodal visual, spatial, and GPS information to perform CoID. In addition, we design a new technique to explicitly address object non-covisibility caused by occlusion and the vehicle's limited field of view. We evaluate our approach in a variety of street environments using a high-fidelity simulation that integrates the CARLA and SUMO simulators. The experimental results show that our approach outperforms the previous approaches and achieves state-of-the- art CoID performance in connected autonomous driving applications. Our work is available at: https://github.com/gaopeng5/DMGM.git.

We suggest the first system that runs real-time semantic segmentation via deep learning on the weak microcomputer Raspberry Pi Zero v2 (whose price was 15) attached to a toy drone. In particular, since the Raspberry Pi weighs less than 16 grams, and its size is half of a credit card, we could easily attach it to the common commercial DJI Tello toy-drone (<15) attached to a toy drone. In particular, since the Raspberry Pi weighs less than 16 grams, and its size is half of a credit card, we could easily attach it to the common commercial DJI Tello toy-drone (<100, <90 grams, 98 \times 92.5\times 41\times 92.5\times 41 mm). The result is an autonomous drone (no laptop nor human in the loop) that can detect and classify objects in real-time from a video stream of an onboard monocular RGB camera (no GPS or LIDAR sensors). The companion videos demonstrate how this Tello drone scans the lab for people (e.g. for the use of firefighters or security forces) and for an empty parking slot outside the lab. Existing deep learning solutions are either much too slow for real-time computation on such IoT devices, or provide results of impractical quality. Our main challenge was to design a system that takes the best of all worlds among numerous combinations of networks, deep learning platforms/frameworks, compression techniques, and compression ratios. To this end, we provide an efficient searching algorithm that aims to find the optimal combination which results in the best tradeoff between the network running time and its accuracy/performance.

Point cloud registration is a crucial technology in the fields of robotics and computer vision. Despite the significant advances in point cloud registration enabled by Transformer-based methods, limitations persist due to indistinct feature extraction, noise sensitivity, and outlier handling. These limitations stem from three factors: (1) the inefficiency of convolutional neural networks (CNNs) to capture global relationships due to their local receptive fields, resulting in extracted features susceptible to noise; (2) the shallow-wide architecture of Transformers, coupled with a lack of positional information, leading to inefficient information interaction and indistinct feature extraction; and (3) the omission of geometrical compatibility leads to ambiguous identification of incorrect correspondences. To overcome these limitations, we propose the Deep Interactive Full Transformer (DIFT) network for point cloud registration, which consists of three key components: (1) a Point Cloud Structure Extractor (PSE) for modeling global relationships and retrieving structural information; (2) a Point Feature Transformer (PFT) for establishing comprehensive associations and directly learning the relative positions between points; and (3) a Geometric Matching-based Correspondence Confidence Evaluation (GMCCE) method for measuring spatial consistency and estimating correspondence confidence. Experimental results on ModelNet40 and 3DMatch datasets demonstrate the superior performance of our proposed method compared to existing state-of-the-art methods. The code for our method is publicly available at https://github.com/CGuangyan-BIT/DIFT.

Recent works in robotic manipulation through reinforcement learning (RL) or imitation learning (IL) have shown potential for tackling a range of tasks e.g., opening a drawer or a cupboard. However, these techniques generalize poorly to unseen objects. We conjecture that this is due to the high-dimensional action space for joint control. In this paper, we take an alternative approach and separate the task of learning ‘what to do’ from ‘how to do it’ i.e., whole-body control. We pose the RL problem as one of determining the skill dynamics for a disembodied virtual manipulator interacting with articulated objects. The whole-body robotic kinematic control is optimized to execute the high-dimensional joint motion to reach the goals in the workspace. It does so by solving a quadratic programming (QP) model with robotic singularity and kinematic constraints. Our experiments on manipulating complex articulated objects show that the proposed approach is more generalizable to unseen objects with large intra-class variations, outperforming previous approaches. The evaluation results indicate that our approach generates more compliant robotic motion and outperforms the pure RL and IL baselines in task success rates. Additional information and videos are available at https://kl-research.github.io/decoupskill.

We consider a robot that answers questions about its environment by traveling to appropriate places and then sensing. Questions are posed as structured queries and may involve conditional or contingent relationships between observable properties. After formulating this problem, and empha-sizing the advantages of exploiting deducible information, we describe how non-trivial knowledge of the world and queries can be given a convenient, concise, unified representation via reduced ordered binary decision diagrams (BDDs). To use these data structures directly for inference and planning, we introduce a new product operation, and generalize the classic dynamic variable reordering techniques to solve planning problems. Also, finally, we evaluate optimizations that exploit locality.

We consider the problem of decentralized multiagent environmental learning through maximizing the joint information gain among a team of agents. Inspired by subsea applications where bandwidth is severely limited, we explicitly consider the challenge of restricted communication between agents. The environment is modeled as a Gaussian process (GP), and the global information gain maximization problem in a GP is a set-valued optimization problem involving all agents' locally acquired data. We develop a decentralized method to solve it based on decomposition of information gain and exchange of limited subsets of data between agents. A key technical novelty of our approach is that we formulate the incentives for information exchange among agents as a submodular set optimization problem in terms of the log-determinant of their local covariance matrices. Numerical experiments on real-world data demonstrate the ability of our algorithm to explore trade-off between objectives. In particular, we demonstrate favorable performance on mapping problems where both decentralized information gathering and limited information exchange are essential.

This paper presents a decentralized multi-agent trajectory planning (MATP) algorithm that guarantees to generate a safe, deadlock-free trajectory in an obstacle-rich environment under a limited communication range. The proposed algorithm utilizes a grid-based multi-agent path planning (MAPP) algorithm for deadlock resolution, and we introduce the subgoal optimization method to make the agent converge to the waypoint generated from the MAPP without deadlock. In addition, the proposed algorithm ensures the feasibility of the optimization problem and collision avoidance by adopting a linear safe corridor (LSC). We verify that the proposed algorithm does not cause a deadlock in both random forests and dense mazes regardless of communication range, and it outperforms our previous work in flight time and distance. We validate the proposed algorithm through a hardware demonstration with ten quadrotors.

We consider the coordinated escort problem, where a decentralised team of supporting robots implicitly assist the mission of higher-value principal robots. The defining challenge is how to evaluate the effect of supporting robots' actions on the principal robots' mission. To capture this effect, we define two novel auxiliary reward functions for supporting robots called satisfaction improvement and satisfaction entropy, which computes the improvement in probability of mission success, or the uncertainty thereof. Given these reward functions, we coordinate the entire team of principal and supporting robots using decentralised cross entropy method (Dec-CEM), a new extension of CEM to multi-agent systems based on the product distribution approximation. In a simulated object avoidance scenario, our planning framework demonstrates up to two-fold improvement in task satisfaction against conventional decoupled information gathering. The significance of our results is to introduce a new family of algorithmic problems that will enable important new practical applications of heterogeneous multi-robot systems.

In this paper, we present a novel Heavy-Tailed Stochastic Policy Gradient (HT-PSG) algorithm to deal with the challenges of sparse rewards in continuous control problems. Sparse rewards are common in continuous control robotics tasks such as manipulation and navigation and make the learning problem hard due to the non-trivial estimation of value functions over the state space. This demands either reward shaping or expert demonstrations for the sparse reward environment. However, obtaining high-quality demonstrations is quite expensive and sometimes even impossible. We propose a heavy-tailed policy parametrization along with a modified momentum-based policy gradient tracking scheme (HT-SPG) to induce a stable exploratory behavior in the algorithm. The proposed algorithm does not require access to expert demonstrations. We test the performance of HT-SPG on various benchmark tasks of continuous control with sparse rewards such as 1D Mario, Pathological Mountain Car, Sparse Pendulum in OpenAI Gym, and Sparse MuJoCo environments (Hopper-v2, Half-Cheetah, Walker-2D). We show consistent performance improvement across all tasks in terms of high average cumulative reward without requiring access to expert demonstrations. We further demonstrate that a navigation policy trained using HT-SPG can be easily transferred into a Clearpath Husky robot to perform real-world navigation tasks.

Recent work has demonstrated the ability of deep reinforcement learning (RL) algorithms to learn complex robotic behaviours in simulation, including in the domain of multi-fingered manipulation. However, such models can be challenging to transfer to the real world due to the gap between simulation and reality. In this paper, we present our techniques to train a) a policy that can perform robust dexterous manipulation on an anthropomorphic robot hand and b) a robust pose estimator suitable for providing reliable real-time information on the state of the object being manipulated. Our policies are trained to adapt to a wide range of conditions in simulation. Consequently, our vision-based policies significantly outperform the best vision policies in the literature on the same reorientation task and are competitive with policies that are given privileged state information via motion capture systems. Our work reaffirms the possibilities of sim-to-real transfer for dexterous manipulation in diverse kinds of hardware and simulator setups, and in our case, with the Allegro Hand and Isaac Gym GPU-based simulation. Furthermore, it opens up possibilities for researchers to achieve such results with commonly-available, affordable robot hands and cameras. Videos of the resulting policy and supplementary information, including experiments and demos, can be found on the website.

Data-driven approaches have shown promising results in modeling and controlling robots, specifically soft and flexible robots where developing physics-based models are more challenging. However, these methods often require a large number of real data, and gathering such data is time-consuming and can damage the robot as well. This paper proposed a novel data-efficient and non-parametric approach to develop a continuous model using a small dataset of real robot demonstrations (only 25 points). To the best of our knowledge, the proposed approach is the most sample-efficient method for soft continuum robot. Furthermore, we employed this model to develop a controller to track arbitrary trajectories in the feasible kinematic space. To show the performance of the proposed approach, a set of trajectory-tracking experiments has been conducted. The results showed that the robot was able to track the references precisely even in presence of external loads (up to 25 grams). Moreover, fine object manipulation experiments were performed to demonstrate the effectiveness of the proposed method in real-world tasks. Finally, we compared its performance with common data-driven approaches in seen/useen-before trajectory tracking scenarios. The results validated that the proposed approach significantly outperformed the existing approaches in unseen-before scenarios and offered similar performance in seen-before scenarios.

In this work we approach the dual optimal reach-safe control problem using sparse approximations of Koopman operator. Matrix approximation of Koopman operator needs to solve a least-squares (LS) problem in the lifted function space, which is computationally intractable for fine discretizations and high dimensions. The state transitional physical meaning of the Koopman operator leads to a sparse LS problem in this space. Leveraging this sparsity, we propose an efficient method to solve the sparse LS problem where we reduce the problem dimension dramatically by formulating the problem using only the non-zero elements in the approximation matrix with known sparsity pattern. The obtained matrix approximation of the operators is then used in a dual optimal reach-safe problem formulation where a linear program with sparse linear constraints naturally appears. We validate our proposed method on various dynamical systems and show that the computation time for operator approximation is greatly reduced with high precision in the solutions.

Simultaneous localization and mapping (SLAM) frameworks for autonomous navigation rely on robust data association to identify loop closures for back-end trajectory optimization. In the case of autonomous underwater vehicles (AUVs) equipped with multibeam echosounders (MBES), data association is particularly challenging due to the scarcity of identifiable landmarks in the seabed, the large drift in deadreckoning navigation estimates to which AUVs are prone and the low resolution characteristic of MBES data. Deep learning solutions to loop closure detection have shown excellent performance on data from more structured environments. However, their transfer to the seabed domain is not immediate and efforts to port them are hindered by the lack of bathymetric datasets. Thus, in this paper we propose a neural network architecture aimed to showcase the potential of adapting such techniques to correspondence matching in bathymetric data. We train our framework on real bathymetry from an AUV mission and evaluate its performance on the tasks of loop closure detection and coarse point cloud alignment. Finally, we show its potential against a more traditional method and release both its implementation and the dataset used.

To make robots accessible to a broad audience, it is critical to endow them with the ability to take universal modes of communication, like commands given in natural language, and extract a concrete desired task specification, defined using a formal language like linear temporal logic (LTL). In this paper, we present a learning-based approach for translating from natural language commands to LTL specifications with very limited human-labeled training data. This is in stark contrast to existing natural-language to LTL translators, which require large human-labeled datasets, often in the form of labeled pairs of LTL formulas and natural language commands, to train the translator. To reduce reliance on human data, our approach generates a large synthetic training dataset through algorithmic generation of LTL formulas, conversion to structured English, and then exploiting the paraphrasing capabilities of modern large language models (LLMs) to synthesize a diverse corpus of natural language commands corresponding to the LTL formu-las. We use this generated data to finetune an LLM and apply a constrained decoding procedure at inference time to ensure the returned LTL formula is syntactically correct. We evaluate our approach on three existing LTL/natural language datasets and show that we can translate natural language commands at 75% accuracy with far less human data (≤12 annotations). Moreover, when training on large human-annotated datasets, our method achieves higher test accuracy (95% on average) than prior work. Finally, we show the translated formulas can be used to plan long-horizon, multi-stage tasks on a 12D quadrotor.

This paper proposes an integration of surrogate modeling and topology to significantly reduce the amount of data required to describe the underlying global dynamics of robot controllers, including closed-box ones. A Gaussian Process (GP), trained with randomized short trajectories over the state-space, acts as a surrogate model for the underlying dynamical system. Then, a combinatorial representation is built and used to describe the dynamics in the form of a directed acyclic graph, known as Morse graph. The Morse graph is able to describe the system's attractors and their corresponding regions of attraction (RoA). Furthermore, a pointwise confidence level of the global dynamics estimation over the entire state space is provided. In contrast to alternatives, the framework does not require estimation of Lyapunov functions, alleviating the need for high prediction accuracy of the GP. The framework is suit-able for data-driven controllers that do not expose an analytical model as long as Lipschitz-continuity is satisfied. The method is compared against established analytical and recent machine learning alternatives for estimating Roas, outperforming them in data efficiency without sacrificing accuracy. Link to code: https://go.rutgers.edu/49hy35en

This paper proposes a probabilistic motion prediction method for long motions. The motion is predicted so that it accomplishes a task from the initial state observed in the given image. While our method evaluates the task achievability by the Energy-Based Model (EBM), previous EBMs are not designed for evaluating the consistency between different domains (i.e., image and motion in our method). Our method seamlessly integrates the image and motion data into the image feature domain by spatially-aligned temporal encoding so that features are extracted along the motion trajectory projected onto the image. Furthermore, this paper also proposes a data-driven motion optimization method, Deep Motion Optimizer (DMO), that works with EBM for motion prediction. Different from previous gradient-based optimizers, our self-supervised DMO alleviates the difficulty of hyper-parameter tuning to avoid local minima. The effectiveness of the proposed method is demonstrated with a variety of experiments with similar SOTA methods.

Modeling and control of high-dimensional, nonlinear robotic systems remains a challenging task. While various model- and learning-based approaches have been proposed to address these challenges, they broadly lack generalizability to different control tasks and rarely preserve the structure of the dynamics. In this work, we propose a new, data-driven approach for extracting control-oriented, low-dimensional models from data using Spectral Submanifold Reduction (SSMR). In contrast to other data-driven methods which fit dynamical models to training trajectories, we identify the dynamics on generic, low-dimensional attractors embedded in the full phase space of the robotic system. This allows us to obtain computationally-tractable models for control which preserve the system's dominant dynamics and better track trajectories radically different from the training data. We demonstrate the superior performance and generalizability of SSMR in dynamic trajectory tracking tasks vis-á-vis the state of the art, including Koopman operator-based approaches.