IROS 2023 论文列表

We study the problem of learning a function that maps context observations (input) to parameters of a submodular function (output). Our motivating case study is a specific type of vehicle routing problem, in which a team of Unmanned Ground Vehicles (UGVs) can serve as mobile charging stations to recharge a team of Unmanned Ground Vehicles (UAVs) that execute persistent monitoring tasks. We want to learn the mapping from observations of UAV task routes and wind field to the parameters of a submodular objective function, which describes the distribution of landing positions of the UAVs. Traditionally, such a learning problem is solved independently as a prediction phase without considering the downstream task optimization phase. However, the loss function used in prediction may be misaligned with our final goal, i.e., a good routing decision. Good performance in the isolated prediction phase does not necessarily lead to good decisions in the downstream routing task. In this paper, we propose a framework that incorporates task optimization as a differentiable layer in the prediction phase. Our framework allows end-to-end training of the prediction model without using engineered intermediate loss that is targeted only at the prediction performance. In the proposed framework, task optimization (submodular maximization) is made differentiable by introducing stochastic perturbations into deterministic algorithms (i.e., stochastic smoothing). We demonstrate the efficacy of the proposed framework using synthetic data. Experimental results of the mobile charging station routing problem show that the proposed framework can result in better routing decisions, e.g. the average number of UAVs recharged increases, compared to the prediction-optimization separate approach.

Aerial tracking with multiple unmanned aerial vehicles (UAVs) has wide potential in various applications. However, the existing works for swarm tracking typically lack the capability of maintaining high target visibility in cluttered environments. To address this deficiency, we present a decentralized planner that maximizes target visibility while ensuring collision-free maneuvers for swarm tracking. In this paper, each drone's tracking performance is first analyzed by a decentralized kinodynamic searching front-end, which renders an optimal guiding path to initialize safe flight corridors and visible sectors. Afterwards, a polynomial trajectory satisfying the corridor constraints is generated by a spatial-temporal optimizer. Inter-vehicle collision and occlusion avoidance are also incorporated into the optimization objectives. The advantages of our methods are verified by extensive benchmark comparisons against other cutting-edge works. Integrated with an autonomous LiDAR-based swarm system, the proposed planner demonstrates its efficiency and robustness in real-world experiments with unknown cluttered surroundings.

Robot swarm is a hot spot in robotic research community. In this paper, we propose a decentralized framework for car-like robotic swarm which is capable of real-time planning in cluttered environments. In this system, path finding is guided by environmental topology information to avoid frequent topological change, and search-based speed planning is leveraged to escape from infeasible initial value's local minima. Then spatial-temporal optimization is employed to generate a safe, smooth and dynamically feasible trajectory. During optimization, the trajectory is discretized by fixed time steps. Penalty is imposed on the signed distance between agents to realize collision avoidance, and differential flatness cooperated with limitation on front steer angle satisfies the non-holonomic constraints. With trajectories broadcast to the wireless network, agents are able to check and prevent potential collisions. We validate the robustness of our system in simulation and real-world experiments. Code will be released as open-source packages.

Deep reinforcement learning (DRL) has seen re-markable success in the control of single robots. However, applying DRL to robot swarms presents significant challenges. A critical challenge is non-stationarity, which occurs when two or more robots update individual or shared policies concurrently, thereby engaging in an interdependent training process with no guarantees of convergence. Circumventing non-stationarity typically involves training the robots with global information about other agents' states and/or actions. In contrast, in this paper we explore how to remove the need for global information. We pose our problem as a Partially Observable Markov Decision Process, due to the absence of global knowledge on other agents. Using collective transport as a testbed scenario, we study two approaches to multi-agent training. In the first, the robots exchange no messages, and are trained to rely on implicit communication through push-and-pull on the object to transport. In the second approach, we introduce Global State Prediction (GSP), a network trained to form a belief over the swarm as a whole and predict its future states. We provide a comprehensive study over four well-known deep reinforcement learning algorithms in environments with obstacles, measuring performance as the successful transport of the object to a goal location within a desired time-frame. Through an ablation study, we show that including GSP boosts performance and increases robustness when compared with methods that use global knowledge.

We present a decentralized connectivity-maintenance algorithm for controlling a group of quadrotor UAVs with limited field of view (FOV) and not sharing a common reference frame for collectively expressing measurements and commands. This is in contrast to the vast majority of previous works on this topic which, instead, make the (simplifying) assumptions of omnidirectional sensing and presence of a common shared frame. For achieving this goal, we design a gradient-based connectivity-maintenance controller able to take into account the presence of a limited FOV. We also propose a novel (to our knowledge) decentralized estimator of the relative orientation among neighboring robots, which is a necessary quantity for correctly implementing the connectivity-maintenance action. We validate the framework in realistic simulations that show the effectiveness of our approach.

Autonomous robotic systems are useful in automating tasks such as inspection and surveying of unknown areas, where speed is often an important factor. In order to effectively reduce the time required to complete missions, an efficient exploration and coordination strategy is needed. In this spirit, this work proposes an approach based on the Monte Carlo Tree Search (MCTS) algorithm to guide robots during exploration missions. Our method first expands a search tree of possible actions from the robot's position towards unknown regions, and then selects the sequence of movements that best drive the exploration process forward with respect to a given reward function. The proposed approach, which is able to balance short- and long-term decision-making, is then extended to accommodate the presence of multiple robots, in a bid to push the efficiency of exploration further. Our method allows for the coordination of the robots' movements in a decentralized manner, relying on point-to-point communication. This results in an efficient strategy, which we refer to as Decentralized Monte Carlo Exploration (DMCE). The experimental results demonstrate that our pipeline outperforms a greedy exploration approach, as well as state-of-the-art planners, with up to 30% reduction in exploration times in a series of real-world maps.

We present a novel approach for fast and reliable policy selection for navigation in partial maps. Leveraging the recent learning-augmented model-based Learning over Subgoals Planning (LSP) abstraction to plan, our robot reuses data collected during navigation to evaluate how well other alternative policies could have performed via a procedure we call offline all-policy replay. Costs from offline alt-policy replay constrain policy selection among the LSP-based policies during deployment, allowing for improvements in convergence speed, cumulative regret and average navigation cost. With only lim-ited prior knowledge about the nature of unseen environments, we achieve at least 67% and as much as 96% improvements on cumulative regret over the baseline bandit approach in our experiments in simulated maze and office-like environments.

We present an implementation of an online op-timization algorithm for hitting a predefined target when returning ping-pong balls with a table tennis robot. The online algorithm optimizes over so-called interception policies, which define the manner in which the robot arm intercepts the ball. In our case, these are composed of the state of the robot arm (position and velocity) at interception time. Gradient information is provided to the optimization algorithm via the mapping from the interception policy to the landing point of the ball on the table, which is approximated with a black-box and a grey-box approach. Our algorithm is applied to a robotic arm with four degrees of freedom that is driven by pneumatic artificial muscles. As a result, the robot arm is able to return the ball onto any predefined target on the table after about 2–5 iterations. We highlight the robustness of our approach by showing rapid convergence with both the black-box and the grey-box gradients. In addition, the small number of iterations required to reach close proximity to the target also underlines the sample efficiency. A demonstration video can be found here: https://youtu.be/VC3KJoCss0k.

This paper presents a data-driven framework to discover underlying dynamics on a scaled F1TENTH vehicle using the Koopman operator linear predictor. Traditionally, a range of white, gray, or black-box models are used to develop controllers for vehicle path tracking. However, these models are constrained to either linearized operational domains, unable to handle significant variability or lose explainability through end-2-end operational settings. The Koopman Extended Dynamic Mode Decomposition (EDMD) linear predictor seeks to utilize data-driven model learning whilst providing benefits like explainability, model analysis and the ability to utilize linear model-based control techniques. Consider a trajectory-tracking problem for our scaled vehicle platform. We collect pose measurements of our F1TENTH car undergoing standard vehicle dynamics benchmark maneuvers with an OptiTrack indoor localization system. Utilizing these uniformly spaced temporal snapshots of the states and control inputs, a data-driven Koopman EDMD model is identified. This model serves as a linear predictor for state propagation, upon which an MPC feedback law is designed to enable trajectory tracking. The prediction and control capabilities of our framework are highlighted through real-time deployment on our scaled vehicle.

Distributionally Robust Optimal Control (DROC) is a technique that enables robust control in a stochastic setting when the true distribution is not known. Traditional DROC approaches require given ambiguity sets or a KL divergence bound to represent the distributional uncertainty. These may not be known a priori and may require hand-crafting. In this paper, we lift this assumption by introducing a data-driven technique for estimating the uncertainty and a bound for the KL divergence. We call this technique D3ROC. To evaluate the effectiveness of our approach, we consider a navigation problem for a car-like robot with unknown noise distributions. The results demonstrate that D3ROC provides robust and efficient control policies that outperform the iterative Linear Quadratic Gaussian (iLQG) control. The results also show the effectiveness of our proposed approach in handling different noise distributions.

Recently, data-driven approaches have brought both opportunities and challenges for Inertial Navigation Systems. In this paper, we propose a novel data-driven method which is composed of cascading orientation and translation estimation with IMU-only measurements. For robust orientation estimation, we combine a CNN-based neural network with an EKF to eliminate orientation errors caused by sensor noises. We additionally propose a hybrid CNN-Transformer-based neural network which exploits both spatial and long-term temporal information to regress accurate translations. Specifically, we conduct detailed evaluations on datasets acquired by iPhone and Android devices. The result demonstrates that our method outperforms state-of-the-art methods in both orientation and translation errors.

This paper presents an online framework for synthesizing agile locomotion for bipedal robots that adapts to unknown environments, modeling errors, and external disturbances. To this end, we leverage step-to-step (S2S) dynamics which has proven effective in realizing dynamic walking on underactuated robots-assuming known dynamics and environments. This paper considers the case of uncertain models and environments and presents a data-driven representation of the S2S dynamics that can be learned via an adaptive control approach that is both data-efficient and easy to implement. The learned S2S controller generates desired discrete foot placement, which is then realized on the full-order dynamics of the bipedal robot by tracking desired outputs synthesized from the given foot placement. The benefits of the proposed approach are twofold. First, it improves the ability of the robot to walk at a given desired velocity when compared to the non-adaptive baseline controller. Second, the data-driven approach enables stable and agile locomotion under the effect of various unknown disturbances: additional unmodeled payload, large robot model errors, external disturbance forces, biased velocity estimation, and sloped terrains. This is demonstrated through in-depth evaluation with a high-fidelity simulation of the bipedal robot Cassie subject to the aforementioned disturbances [1].

This paper develops a data-based moving horizon estimation (MHE) method for agile quadrotors. Accurate state estimation of the system is paramount for precise trajectory control for agile quadrotors; however, the high level of aerodynamic forces experienced by the quadrotors during high-speed flights make this task extremely challenging. These complex turbulent effects are difficult to model and the unmodelled dynamics introduce inaccuracies in the state estimation. In this work, we propose a method to model these aerodynamic effects using Gaussian Processes which we integrate into the MHE to achieve efficient and accurate state estimation with minimal computational burden. Through extensive simulation and experimental studies, this method has demonstrated significant improvement in state estimation performance displaying superior robustness to poor state measurements.

The range of robot activities is expanding from industries with fixed environments to diverse and changing environments, such as nursing care support and daily life support. In particular, autonomous construction of robots that are personalized for each user and task is required. Therefore, we develop an actuator module that can be reconfigured to various link configurations, can carry heavy objects using a locking mechanism, and can be easily operated by human teaching using a releasing mechanism. Given multiple target coordinates, a modular robot configuration that satisfies these coordinates and minimizes the required torque is automatically generated by Tree-structured Parzen Estimator (TPE), a type of black-box optimization. Based on the obtained results, we show that the robot can be reconfigured to perform various functions such as moving monitors and lights, serving food, and so on.

Autonomous underwater gliders (AUGs) are capable of traversing basin scale distances but lack sufficient localization accuracy to operate without periodic surfacing to obtain GPS fixes and constrain localization drift. Conventionally, AUGs use a dynamic flight model with depth averaged current correction (DACC) to dead-reckon their position while subsurface. However, these flight models become unstable at shallow pitch angles and DACC is inaccurate in dynamic and highly sheared water column currents. We present a method and preliminary results from field trials for improved real-time AUG localization using a Doppler Velocity Logger to estimate vehicle velocity and dynamically profile water column currents. This improved localization reduces the need for periodic surfacing, while independence from a dynamic flight model makes it particularly suited for shallow-yo profile missions.

We present an algorithm for safe robot navigation in complex dynamic environments using a variant of model predictive equilibrium point control. We use an optimization formulation to navigate robots gracefully in dynamic environments by optimizing over a trajectory cost function at each timestep. We present a novel trajectory cost formulation that significantly reduces conservative and deadlocking behaviors and generates smooth trajectories. In particular, we propose a new collision probability function that effectively captures the risk associated with a given configuration and the time to avoid collisions based on the velocity direction. Moreover, we propose a terminal state cost based on the expected time-to-goal and time-to-collision values that helps in avoiding trajectories that could result in deadlock. We evaluate our cost formulation in multiple simulated scenarios, including narrow corridors with dynamic obstacles, and observe significantly improved navigation behavior and reduced deadlocks as compared to prior methods.

The performance of local feature descriptors degrades in the presence of large rotation variations. To address this issue, we present an efficient approach to learning rotation invariant descriptors. Specifically, we propose Rotated Kernel Fusion (RKF) which imposes rotations on the convolution kernel to improve the inherent nature of CNN. Since RKF can be processed by the subsequent re-parameterization, no extra computational costs will be introduced in the inference stage. Moreover, we present Multi-oriented Feature Aggregation (MOFA) which aggregates features extracted from multiple rotated versions of the input image and can provide auxiliary knowledge for the training of RKF by leveraging the distillation strategy. We refer to the distilled RKF model as DRKF. Besides the evaluation on a rotation-augmented version of the public dataset HPatches, we also contribute a new dataset named DiverseBEV which is collected during the drone's flight and consists of bird's eye view images with large viewpoint changes and camera rotations. Extensive experiments show that our method can outperform other state-of-the-art techniques when exposed to large rotation variations.

Category-level object pose estimation involves estimating the 6D pose and the 3D metric size of objects from predetermined categories. While recent approaches take categorical shape prior information as reference to improve pose estimation accuracy, the single-stage network design and training manner lead to sub-optimal performance since there are two distinct tasks in the pipeline. In this paper, the advantage of two-stage pipeline over single-stage design is discussed. To this end, we propose a two-stage deformation-and-registration pipeline called DR-Pose, which consists of completion-aided deformation stage and scaled registration stage. The first stage uses a point cloud completion method to generate unseen parts of target object, guiding subsequent deformation on the shape prior. In the second stage, a novel registration network is designed to extract pose-sensitive features and predict the representation of object partial point cloud in canonical space based on the deformation results from the first stage. DR-Pose produces superior results to the state-of-the-art shape prior-based methods on both CAMERA25 and REAL275 benchmarks. Codes are available at https://github.com/Zray26/DR-Pose.git.

This work presents a dataset of human-operated robot motion to be used within the context of assistive robotics and assorted fields, such as learning from demonstrations, machine-learning based robot control, and activity recognition. The data consists of individual sequences of intentional robot motion performing a task in an environment of daily living. There are 2973 sequences generated in a high-resolution simulation and 986 sequences performed in reality, totaling to 1.16 M datapoints. The data includes labels for the robot's pose, motion and activity. This paper also provides data augmentation methods and a detailed dataset analysis as well as simple models trained on the dataset as a baseline for future research. The dataset can be downloaded free-of-charge at https://www.kaggle.com/f371xx/dormadl.