ICRA 2023 Papers

Objects rarely sit in isolation in human environments. As such, we'd like our robots to reason about how multiple objects relate to one another and how those relations may change as the robot interacts with the world. To this end, we propose a novel graph neural network framework for multi-object manipulation to predict how inter-object relations change given robot actions. Our model operates on partial-view point clouds and can reason about multiple objects dynamically interacting during the manipulation, By learning a dynamics model in a learned latent graph embedding space, our model enables multi-step planning to reach target goal relations. We show our model trained purely in simulation transfers well to the real world. Our planner enables the robot to rearrange a variable number of objects with a range of shapes and sizes using both push and pick-and-place skills.

Real-world manipulation problems in heavy clutter require robots to reason about potential contacts with objects in the environment. We focus on pick-and-place style tasks to retrieve a target object from a shelf where some ‘movable’ objects must be rearranged in order to solve the task. In particular, our motivation is to allow the robot to reason over and consider non-prehensile rearrangement actions that lead to complex robot-object and object-object interactions where multiple objects might be moved by the robot simultaneously, and objects might tilt, lean on each other, or topple. To support this, we query a physics-based simulator to forward simulate these interaction dynamics which makes action evaluation during planning computationally very expensive. To make the planner tractable, we establish a connection between the domain of Manipulation Among Movable Objects and Multi-Agent Pathfinding that lets us decompose the problem into two phases our M4M algorithm iterates over. First we solve a multi-agent planning problem that reasons about the configurations of movable objects but does not forward simulate a physics model. Next, an arm motion planning problem is solved that uses a physics-based simulator but does not search over possible configurations of movable objects. We run simulated and real-world experiments with the PR2 robot and compare against relevant baseline algorithms. Our results highlight that M4M generates complex 3D interactions, and solves at least twice as many problems as the baselines with competitive performance.

Assembly Sequence Planning (ASP) is the essential process for modern manufacturing, proven to be NP-complete thus its effective and efficient solution has been a challenge for researchers in the field. In this paper, we present a graph-transformer based framework for the ASP problem which is trained and demonstrated on a self-collected ASP database. The ASP database contains a self-collected set of LEGO models. The LEGO model is abstracted to a heterogeneous graph structure after a thorough analysis of the original structure and feature extraction. The ground truth assembly sequence is first generated by brute-force search and then adjusted manually to be in line with human rational habits. Based on this self-collected ASP dataset, we propose a heterogeneous graph-transformer framework to learn the latent rules for assembly planning. We evaluated the proposed framework in a series of experiments. The results show that the similarity of the predicted and ground truth sequences can reach 0.44, a medium correlation measured by Kendall's τ. Meanwhile, we compared the different effects of node features and edge features and generated a feasible and reasonable assembly sequence as a benchmark for further research. Our dataset and code are available on: htps://github.com/AIR-DISCOVER/ICRA_ASP.

Place recognition is a critical and challenging task for mobile robots, aiming to retrieve an image captured at the same place as a query image from a database. Existing methods tend to fail while robots move autonomously under occlusion (e.g., car, bus, truck) and changing appearance (e.g., illumination changes, seasonal variation). Because they encode the image into only one code, entangling place features with appearance and occlusion features. To overcome this limitation, we propose PROCA, an unsupervised approach to decompose the image representation into three codes: a place code used as a descriptor to retrieve images, an appearance code that captures appearance properties, and an occlusion code that encodes occlusion content. Extensive experiments show that our model outperforms the state-of-the-art methods. Our code and data are available at https://github.com/rover-xingyu/PROCA.

This paper presents a robotic in-hand manipulation technique that can be applied to pick an object too large to grasp in a prehensile manner, by taking advantage of its contact interactions with a curved, passive end-effector, and two flat support surfaces. First, the object is tilted up while being held between the end-effector and the supports. Then, the end-effector is tucked into the gap underneath the object, which is formed by tilting, in order to obtain a grasp against gravity. In this paper, we first examine the mechanics of tilting to understand the different ways in which the object can be initially tilted. We then present a strategy to tilt up the object in a secure manner. Finally, we demonstrate successful picking of objects of various size and geometry using our technique through a set of experiments performed with a custom-made robotic device and a conventional robot arm. Our experiment results show that object picking can be performed reliably with our method using simple hardware and control, and when possible, with appropriate fixture design.

The choice of a grasp plays a critical role in the success of downstream manipulation tasks. Consider a task of placing an object in a cluttered scene; the majority of possible grasps may not be suitable for the desired placement. In this paper, we study the synergy between the picking and placing of an object in a cluttered scene to develop an algorithm for task-aware grasp estimation. We present an object-centric action space that encodes the relationship between the geometry of the placement scene and the object to be placed in order to provide placement affordance maps directly from perspective views of the placement scene. This action space enables the computation of a one-to-one mapping between the placement and picking actions allowing the robot to generate a diverse set of pick-and-place proposals and to optimize for a grasp under other task constraints such as robot kinematics and collision avoidance. With experiments both in simulation and on a real robot we demonstrate that with our method, the robot is able to successfully complete the task of placement-aware grasping with over 89 % accuracy in such a way that generalizes to novel objects and scenes.

Combining task and motion planning (TAMP) is crucial for intelligent robots to perform complex and long-horizon tasks. In TAMP, many approaches generally employ Monte-Carlo tree search (MCTS) with upper confidence bound (UCB) for task planning to handle exploration-exploitation trade-off and find globally optimal solutions. However, since UCB basically considers the estimation error caused by noise, the error caused by insufficient optimization of the sub-tree is not represented. Hence, UCB-based approaches have the disadvantage of not exploring underestimated sub-trees. To alleviate this issue, we propose a novel tree search method using perturbation-based best-arm identification (PBAI). We theoretically prove the bound of the simple regret of our method and empirically verify that PBAI finds the optimal task plans faster and more efficiently than the existing algorithms. The source code of our proposed algorithm is available at https://github.com/jdj2261/pytamp.

The recent development of high-precision subsea optical scanners allows for 3D keypoint detectors and feature descriptors to be leveraged on point cloud scans from subsea environments. However, the literature lacks a comprehensive survey to identify the best combination of detectors and descriptors to be used in these challenging and novel environments. This paper aims to identify the best detector/descriptor pair using a challenging field dataset collected using a commercial underwater laser scanner. Furthermore, studies have shown that incorporating texture information to extend geometric features adds robustness to feature matching on synthetic datasets. This paper also proposes a novel method of fusing images with underwater laser scans to produce coloured point clouds, which are used to study the effectiveness of 6D point cloud descriptors.

We present a novel approach to interactive 3D object perception for robots. Unlike previous perception algorithms that rely on known object models or a large amount of annotated training data, we propose a poking-based approach that automatically discovers and reconstructs 3D objects. The poking process not only enables the robot to discover unseen 3D objects but also produces multi-view observations for 3D reconstruction of the objects. The reconstructed objects are then memorized by neural networks with regular supervised learning and can be recognized in new test images. The experiments on real-world data show that our approach could unsupervisedly discover and reconstruct unseen 3D objects with high quality, and facilitate real-world applications such as robotic grasping. The code and supplementary materials are available at the project page: https://zju3dv.github.io/poking_perception/.

Accurate understanding and prediction of human behaviors are critical prerequisites for autonomous vehicles, especially in highly dynamic and interactive scenarios such as intersections in dense urban areas. In this work, we aim at identifying crossing pedestrians and predicting their future trajectories. To achieve these goals, we not only need the context information of road geometry and other traffic participants but also need fine-grained information of the human pose, motion and activity, which can be inferred from human keypoints. In this paper, we propose a novel multi-task learning framework for pedestrian crossing action recognition and trajectory pre-diction, which utilizes 3D human keypoints extracted from raw sensor data to capture rich information on human pose and activity. Moreover, we propose to apply two auxiliary tasks and contrastive learning to enable auxiliary supervisions to improve the learned keypoints representation, which further enhances the performance of major tasks. We validate our approach on a large-scale in-house dataset, as well as a public benchmark dataset, and show that our approach achieves state-of-the-art performance on a wide range of evaluation metrics. The effectiveness of each model component is validated in a detailed ablation study.

Predicting pedestrian behavior is a crucial task for intelligent driving systems. Accurate predictions require a deep understanding of various contextual elements that could impact the way pedestrians behave. To address this challenge, we propose a novel framework that relies on different data modalities to predict future trajectories and crossing actions of pedestrians from an egocentric perspective. Specifically, our model utilizes a cross-modal Transformer architecture to capture dependencies between different data types. The output of the Transformer is augmented with representations of interactions between pedestrians and other traffic agents conditioned on the pedestrian and ego-vehicle dynamics that are generated via a semantic attentive interaction module. Lastly, the context encodings are fed into a multi-stream decoder framework using a gated-shared network. We evaluate our algorithm on public pedestrian behavior benchmarks, PIE and JAAD, and show that our model improves state-of-the-art in trajectory and action prediction by up to 22% and 13% respectively on various metrics. The advantages of the proposed components are investigated via extensive ablation studies.

In this paper, we study a navigation problem where a mobile robot needs to locate a mmWave wireless signal. Using the directionality properties of the signal, we propose an estimation and path planning algorithm that can efficiently navigate in cluttered indoor environments. We formulate Extended Kalman filters for emitter location estimation in cases where the signal is received in line-of-sight or after reflections. We then propose to plan motion trajectories based on belief-space dynamics in order to minimize the uncertainty of the position estimates. The associated non-linear optimization problem is solved by a state-of-the-art constrained iLQR solver. In particular, we propose a method that can handle a large number of obstacles (∼ 300) with reasonable computation times. We validate the approach in an extensive set of simulations. We show that our estimators can help increase navigation success rate and that planning to reduce estimation uncertainty can improve the overall task completion speed.

This paper proposes a decentralized passive impedance control scheme for collaborative grasping using under-actuated aerial manipulators (AMs). The AM system is formulated, using a proper coordinate transformation, as an inertially decoupled dynamics with which a passivity-based control design is conducted. Since the interaction for grasping can be interpreted as a feedback interconnection of passive systems, an arbitrary number of AMs can be modularly combined, leading to a decentralized control scheme. Another interesting consequence of the passivity property is that the AMs automatically converge to a certain configuration to accomplish the grasping. Collaborative grasping using 10 AMs is presented in simulation.

Robotic end-effectors have been developed for various applications. Most of them are driven by electric or pneumatic actuator/actuators, which usually make the end-effector bulky and vulnerable due to the external cables and air tubes. In this study, we propose a novel passive robotic gripper with a locking mechanism that does not require any actuators. Locking and unlocking of the gripper fingers are performed through contact with external environment, such as ground, table, and conveyor. To facilitate gripper design, modeling of the deformed finger shape was conducted, and experimental validation was performed. A robotic gripper with eight such passive fingers were fabricated using 3D printer. Experiments were conducted to investigate the grasping capacities in terms of object size and weight. We found that the larger the object, the greater the weight capacity of the gripper, which increased significantly when the object exceeded a certain size. In addition, experiments on grasping various food products were carried out and results suggested that the proposed gripper could grasp objects with complex shapes and soft fragile properties, but damages were caused on very fragile objects due to the rigid structure of the gripper.

Hamilton-Jacobi (HJ) reachability analysis is a powerful tool for analyzing the safety of autonomous systems. However, the provided safety assurances are often predicated on the assumption that once deployed, the system or its environment does not evolve. Online, however, an autonomous system might experience changes in system dynamics, control authority, external disturbances, and/or the surrounding environment, requiring updated safety assurances. Rather than restarting the safety analysis from scratch, which can be timeconsuming and often intractable to perform online, we propose to compute parameter-conditioned reachable sets. Assuming expected system and environment changes can be parameterized, we treat these parameters as virtual states in the system and leverage recent advances in high-dimensional reachability analysis to solve the corresponding reachability problem offline. This results in a family of reachable sets that is parameterized by the environment and system factors. Online, as these factors change, the system can simply query the corresponding safety function from this family to ensure system safety, enabling a real-time update of the safety assurances. Through various simulation studies, we demonstrate the capability of our approach in maintaining system safety despite the system and environment evolution.

Sampling-based motion planning algorithms have been continuously developed for more than two decades. Apart from mobile robots, they are also widely used in manipulator motion planning. Hence, these methods play a key role in collaborative and shared workspaces. Despite numerous improvements, their performance can highly vary depending on the chosen parameter setting. The optimal parameters depend on numerous factors such as the start state, the goal state and the complexity of the environment. Practitioners usually choose these values using their experience and tedious trial and error experiments. To address this problem, recent works combine hyperparameter optimization methods with motion planning. They show that tuning the planner's parameters can lead to shorter planning times and lower costs. It is not clear, however, how well such approaches generalize to a diverse set of planning problems that include narrow passages as well as barely cluttered environments. In this work, we analyze optimized planner settings for a large set of diverse planning problems. We then provide insights into the connection between the characteristics of the planning problem and the optimal parameters. As a result, we provide a list of recommended parameters for various use-cases. Our experiments are based on a novel motion planning benchmark for manipulators which we provide at https://mytuc.org/rybj.

Reinforcement learning (RL) is an agent-based approach for teaching robots to navigate within the physical world. Gathering data for RL is known to be a laborious task, and real-world experiments can be risky. Simulators facilitate the collection of training data in a quicker and more cost-effective manner. However, RL frequently requires a significant number of simulation steps for an agent to become skilful at simple tasks. This is a prevalent issue within the field of RL-based visual quadrotor navigation where state dimensions are typically very large and dynamic models are complex. Furthermore, rendering images and obtaining physical properties of the agent can be computationally expensive. To solve this, we present a simulation framework, built on AirSim, which provides efficient parallel training. Building on this framework, Ape-X is modified to incorporate parallel training of AirSim environments to make use of numerous networked computers. Through experiments we were able to achieve a reduction in training time from 3.9 hours to 11 minutes, for a toy problem, using the aforementioned framework and a total of 74 agents and two networked computers. Further details including a github repo and videos about our project, PRL4AirSim, can be found at https://sites.google.com/view/prl4airsim/home

We present a parallelized optimization method based on fast Neural Radiance Fields (NeRF) for estimating 6-DoF pose of a camera with respect to an object or scene. Given a single observed RGB image of the target, we can predict the translation and rotation of the camera by minimizing the residual between pixels rendered from a fast NeRF model and pixels in the observed image. We integrate a momentum-based camera extrinsic optimization procedure into Instant Neural Graphics Primitives, a recent exceptionally fast NeRF implementation. By introducing parallel Monte Carlo sampling into the pose estimation task, our method overcomes local minima and improves efficiency in a more extensive search space. We also show the importance of adopting a more robust pixel-based loss function to reduce error. Experiments demonstrate that our method can achieve improved generalization and robustness on both synthetic and real-world benchmarks.

Invariant Extended Kalman Filter (IEKF) has been successfully applied in Visual-inertial Odometry (VIO) as an advanced achievement of Kalman filter, showing great potential in sensor fusion. In this paper, we propose partial IEKF (PIEKF), which only incorporates rotation-velocity state into the Lie group structure and apply it for Visual-Inertial-Wheel Odometry (VIWO) to improve positioning accuracy and consistency. Specifically, we derive the rotation-velocity measurement model, which combines wheel measurements with kinematic constraints. The model circumvents the wheel odometer's 3D integration and covariance propagation, which is essential for filter consistency. And a plane constraint is also introduced to enhance the position accuracy. A dynamic outlier detection method is adopted, leveraging the velocity state output. Through the simulation and real-world test, we validate the effectiveness of our approach, which outperforms the standard Multi-State Constraint Kalman Filter (MSCKF) based VIWO in consistency and accuracy.