ICRA 2022 论文列表

Quadrupedal landing is a complex process involving large impacts, elaborate contact transitions, and is a crucial recovery behavior observed in many biological animals. This work presents a real-time, optimal landing controller that is free of pre-specified contact schedules. The controller determines optimal touchdown postures and reaction force profiles and is able to recover from a variety of falling configurations. The quadrupedal platform used, the MIT Mini Cheetah, recovered safely from drops of up to 8 m in simulation, as well as from a range of orientations and planar velocities. The controller is also tested on hardware, successfully recovering from drops of up to 2 m.

This work proposes a robotic pipeline for picking and constrained placement of objects without geometric shape priors. Compared to recent efforts developed for similar tasks, where every object was assumed to be novel, the proposed system recognizes previously manipulated objects and per-forms online model reconstruction and reuse. Over a lifelong manipulation process, the system keeps learning features of objects it has interacted with and updates their reconstructed models. Whenever an instance of a previously manipulated object reappears, the system aims to first recognize it and then register its previously reconstructed model given the current observation. This step greatly reduces object shape uncertainty allowing the system to even reason for parts of objects, which are currently not observable. This also results in better manipulation efficiency as it reduces the need for active perception of the target object during manipulation. To get a reusable reconstructed model, the proposed pipeline adopts: i) TSDF for object representation, and ii) a variant of the standard particle filter algorithm for pose estimation and tracking of the partial object model. Furthermore, an effective way to construct and maintain a dataset of manipulated objects is presented. A sequence of real-world manipulation experiments is performed. They show how future manipulation tasks become more effective and efficient by reusing reconstructed models of previously manipulated objects, which were generated during their prior manipulation, instead of treating objects as novel every time.

This paper presents a Non-Linear Model Predictive Controller for humanoid robot locomotion with online step adjustment capabilities. The proposed controller considers the Centroidal Dynamics of the system to compute the desired contact forces and torques and contact locations. Differently from bipedal walking architectures based on simplified models, the presented approach considers the reduced centroidal model, thus allowing the robot to perform highly dynamic movements while keeping the control problem still treatable online. We show that the proposed controller can automatically adjust the contact location both in single and double support phases. The overall approach is then tested with a simulation of one-leg and two-leg systems performing jumping and running tasks, respectively. We finally validate the proposed controller on the position-controlled Humanoid Robot iCub. Results show that the proposed strategy prevents the robot from falling while walking and pushed with external forces up to 40 Newton for 1 second applied at the robot arm.

Gait generation frameworks for humanoid robots typically assume a constant centroidal angular momentum (CAM) throughout the walking cycle, which induces undesirable contact torques in the feet and results in performance degradation. In this work, we present a novel algorithm to learn the CAM online and include the obtained knowledge within the closed-form solutions of the Divergent Component of Motion (DCM) locomotion framework. To ensure a reduction of the contact torques at the desired center of pressure position, a CAM trajectory is generated and explicitly tracked by a whole-body controller. Experiments with the humanoid robot TORO demonstrate that the proposed method significantly increases the maximum step length and walking speed during locomotion.

Wearable gait assistive devices have been im-proved by soft robotics in terms of safety and physical burden of the wearer. Currently, electrical sensors and computers around the wearer are bottlenecks in enhancing the wearer's activities. In this paper, we present a wearable gait assistive system using soft pneumatic system for all the actuation, sensing, and computation. Pneumatic artifical muscles (PAMs) on the thighs were used to generate assistance force, and thin PAMs on the whole legs were used for sensing the wearer's motion. All thin PAMs were connected to each other by tubes, and the pressure response in the tubes were exploited to compute the wearer's motion in manner of physical reservoir computing. The left thigh angular velocity estimated from the reservoir response was used for control the PAMs for actuation. Our experiment with the use of gait assistance showed that the system worked correctly. This paper shows that the pneumatic analog computation system for sensing soft body can help the functionality of soft wearable assistive devices.

In order to obtain a high-precision contact model that can properly describe the target soft tissue, this paper proposes a hybrid soft contact model based on a clustering algorithm ART-II, which selects the most suitable soft contact model according to the surgical environment. The least-square method is used to identify the parameters of the model online. In the experiments, different parts of animal tissues were used as the experimental objects. The hybrid model was used to identify and switch for the most appropriate soft contact model when dealing with a certain type of animal tissue. The performance of the hybrid model on force estimation was compared with several individual soft contact models. The results showed that the estimated/reconstructed force of the hybrid model was closer to the ground truth measured by the force sensor. In addition, a new reference soft contact model has been purposely added online to verify the expandability of the hybrid model.

Cell fusion has been widely applied in scientific research for cancer immunotherapy, antibody production, and nuclear reprogramming of somatic cells, and therefore the cell fusion technique that enable us to precisely control the fusion process with high throughput manner has been desired. Here, we present a novel microfluidic method for automatic cell pairing by microdroplets, separation of droplets containing cells, and electrofusion of cells inside a droplet. The proposed microfluidic device mainly composed of three sequential function parts for (i) encapsulation of cells into a droplet by microfluidic droplet generator, (ii) separation of droplets containing cells from empty droplets through a micropillar array, and (iii) electrofusion of cells inside the droplets by applying a voltage during the droplet passing over the pair of electrodes. In the microfluidic device, cell-encapsulated and empty droplets were generated at the upstream cross-junction; they then entered the micropillar array, separating the cell-encapsulated droplets from empty droplets continuously. After separation, they passed over the electrode pairs, and were collected the outside of the microchannel. This continuous process for cell fusion would enable us to observe and isolate the target fused cells for cell analysis.

Hyperdimensional computing (HDC) is a brain-inspired computing paradigm that operates on pseudo-random hypervectors, an information-rich, hardware-efficient representation that is robust to noise and facilitates learning with limited training data. This work explores how robot navigation tasks can leverage the high-capacity hypervector representation to enable behavioral prioritization through a weighted encoding of heterogeneous sensor information. Experiments over 100 trials in each of the 100 randomly generated obstacle maps demonstrate that the proposed weighted sensor encoding scheme boosts the success rate of the navigation task by over 30% compared to an unweighted sensor encoding. A hybrid scheme using the HDC weighted scheme at the input of a deep feed-forward neural network achieves the highest success rate. The hybrid scheme furthermore is more robust when reducing the HDC dimension by 50%. However, the simple HDC implementation remains the most hardware efficient, making it desirable for resource-constrained systems.

Inverse Optimal Control (IOC) is a popular method for human motion analysis. In the context of these methods it is necessary to pay attention to the reliability of the results. This paper proposes an approach based on the evaluation of Karush-Kuhn-Tucker conditions relying on a complete analysis with Singular Value Decomposition and provides a detailed analysis of reliability. With respect to a ground truth, our simulations illustrate how the proposed method analyzes the reliability of the resolution. After introducing a clear methodology, the properties of the matrices are studied with different noise levels and different experimental models and conditions. We show how to implement the method, step by step, by explaining the numerical difficulties encountered during the resolution and thus how to make the results of the IOC problem reliable.

Finger-gaiting manipulation is an important skill to achieve large-angle in-hand re-orientation of objects. However, achieving these gaits with arbitrary orientations of the hand is challenging due to the unstable nature of the task. In this work, we use model-free reinforcement learning (RL) to learn finger-gaiting only via precision grasps and demonstrate finger-gaiting for rotation about an axis using only on-board proprioceptive and tactile feedback. To tackle the inherent instability of precision grasping, we propose the use of initial state distributions that enable effective exploration of the state space. Our method can learn finger gaiting with better sample complexity than the state-of-the-art. The policies we obtain are robust to noise and perturbations, and transfer to novel objects. Videos can be found at https://roamlab.github.io/learnfg/

This paper studies the distributed multi-robot constrained navigation problem. While the multi-robot collision avoidance has been extensively studied in the literature with safety being the primary focus, the individual robot's destination convergence is not necessarily guaranteed. In particular, robots may get stuck in the local equilibria or periodic orbits of the multi-robot system, some of which are practically known as the deadlock and the livelock behaviors. Inspired by the combination of Control Lyapunov Function (CLF) and Control Barrier Function (CBF) for the nonlinear system's constrained stabilization, the authors present a guaranteed safe feedback control policy with improved convergence performance. The proposed Parametric CLF (PCLF) scheme adaptively determines the appropriate CLF parameterization within the in-stantaneous feasible action space. The algorithm also induces a conditional global asymptotic convergence guarantee for multi-robot system of single-integrator dynamics, and is empirically effective for nonlinear nonholonomic vehicle model. Empiri-cally, the proposed PCLF-CBF framework exhibits superior performance than state-of-the-art methods, including its de-generated counterpart of various CLF-CBF solutions.

The problem of combinatorial filter reduction arises from resource optimization in robots; it is one specific way in which automation can help to achieve minimalism, to build better robots. This paper contributes a new definition of filter minimization that is broader than its antecedents, allowing filters (input, output, or both) to be nondeterministic. This changes the problem considerably. Nondeterministic filters may re-use states to obtain more ‘behavior’ per vertex. We show that the gap in size can be significant (larger than polyno-mial), suggesting such cases will generally be more challenging than deterministic problems. Indeed, this is supported by the core complexity result established in this paper: producing nondeterministic minimizers is PSPACE-hard. The hardness separation for minimization existing between deterministic filter and automata, thus, fails to hold for the nondeterministic case.

Presented is a new 10-mm diameter wrist designed for robotic surgery. Featuring greater dexterity than current designs, entirely new procedures may be possible. An innovative, parallel mechanism, it offers 180 degrees of singularity-free pitch/yaw motion. The wrist is also a new form of high angulation, constant velocity, universal joint and is capable of continuous 360 degrees of roll rotation at any angle to enable drilling, reaming, milling, filing, deburring, or inserting fasteners. Z-axis motion is incorporated into the wrist itself, a unique feature. In addition, a gripper, scissors, stapler or other open/close device may be added. A thru-hole allows for passing wires, fiber optics, or flexible tubes. Stainless steel rotary shafts drive the wrist and also enable force reflection useful for providing haptic feedback to the operator. Eliminating difficult to sterilize woven steel cables/pulleys and bearings, the wrist may be sterilized in an autoclave. A simple construction, the wrist lends itself to CNC machining and automated assembly.

Robot grasping and manipulation allow robots to interact with their environments and execute a plethora of complex tasks that require increased dexterity (e.g., open a door, push buttons, collect and transpose objects, etc.). Collecting data of such activities is of paramount importance as it allows roboticists to create new methods and models that will facilitate the execution of sophisticated tasks. In this paper, we propose new wearable, lightweight, low-cost human machine interfaces that improve the efficiency of the data collection process for both robotic grasping and manipulation by offering intuitive and simplified control of the employed robotic grippers and hands. In particular, two different types of interfaces are proposed: i) a handle-based forearm stabilized interface that uses a waist-linkage system to provide weight support for bulky and heavy robotic end-effectors and ii) a palm-mounted interface that can accommodate smaller and lightweight grippers and hands, offering more agility in the control and positioning of these devices. Both interfaces are equipped with appropriate sliders, joysticks, and buttons that facilitate the control of the multiple degrees of freedom of the employed end-effectors and appropriate cameras that allow for object detection, identification, and object pose estimation.

Robots excel at avoiding obstacles but struggle to traverse complex 3-D terrain with cluttered large obstacles. By contrast, insects like cockroaches excel at doing so. Recent research in our lab elucidated how locomotor transitions emerge from locomotor-environment interaction for diverse locomotor challenges abstracted from complex 3-D terrain and the strategies to overcome them. Here we built on these fundamental insights to develop a cockroach-inspired legged robot, Omni-Roach, that integrated these strategies to achieve multi-modal locomotion and provide a robophysical model to study the tradeoff between multi-functionality and performance. The robot was based on the RHex design with six compliant legs and featured a rounded body with two wings that can open and a tail with pitch and yaw degrees of freedom. After two development and testing iterations, our robot was capable of overcoming all loco-motor challenges with a high performance and success rate. It traversed cluttered rigid pillars only 1.1× robot body width apart, a 2.5× hip height bump, a 0.75× body length gap, densely cluttered flexible beams only 65% body width apart, and self-righted within 4 seconds. Systematic beam traversal experiments further revealed that a downward-pointing tail oscillating laterally helps roll the body into beam gaps and break frictional and interlocking contact to traverse. Our work highlights the usefulness of multi-functional appendages and exaptation for large obstacle traversal.

Reinforcement learning (RL) can in principle let robots automatically adapt to new tasks, but current RL methods require a large number of trials to accomplish this. In this paper, we tackle rapid adaptation to new tasks through the framework of meta-learning, which utilizes past tasks to learn to adapt with a specific focus on industrial insertion tasks. Fast adaptation is crucial because prohibitively large number of on-robot trials will potentially damage hardware pieces. Additionally, effective adaptation is also feasible in that experience among different insertion applications can be largely leveraged by each other. In this setting, we address two specific challenges when applying meta-learning. First, conventional meta-RL algorithms require lengthy online meta-training. We show that this can be replaced with appropriately chosen offline data, resulting in an offline meta- RL method that only requires demonstrations and trials from each of the prior tasks, without the need to run costly meta-RL procedures online. Second, meta-RL methods can fail to generalize to new tasks that are too different from those seen at meta-training time, which poses a particular challenge in industrial applications, where high success rates are critical. We address this by combining contextual meta-learning with direct online finetuning: if the new task is similar to those seen in the prior data, then the contextual meta-learner adapts immediately, and if it is too different, it gradually adapts through finetuning. We show that our approach is able to quickly adapt to a variety of different insertion tasks, with a success rate of 100% using only a fraction of the samples needed for learning the tasks from scratch. Experiment videos and details are available at //sites.google.com/view/offline-metarl-insertion.https:

We consider real-world reinforcement learning (RL) of robotic manipulation tasks that involve both visuomotor skills and contact-rich skills. We aim to train a policy that maps multimodal sensory observations (vision and force) to a manipulator's joint velocities under practical considerations. We propose to use offline samples to learn a set of general value functions (GVFs) that make counterfactual predictions from the visual inputs. We show that combining the offline learned counterfactual predictions with force feedbacks in online policy learning allows efficient reinforcement learning given only a terminal (success/failure) reward. We argue that the learned counterfactual predictions form a compact and informative representation that enables sample efficiency and provides auxiliary reward signals that guide online explorations towards contact-rich states. Various experiments in simulation and real-world settings were performed for evaluation. Recordings of the real-world robot training can be found via https://sites.google.com/view/realrl.

Applying reinforcement learning (RL) methods on robots typically involves training a policy in simulation and deploying it on a robot in the real world. Because of the model mismatch between the real world and the simulator, RL agents deployed in this manner tend to perform suboptimally. To tackle this problem, researchers have developed robust policy learning algorithms that rely on synthetic noise disturbances. However, such methods do not guarantee performance in the target environment. We propose a convex risk minimization algorithm to estimate the model mismatch between the simulator and the target domain using trajectory data from both environments. We show that this estimator can be used along with the simulator to evaluate performance of an RL agents in the target domain, effectively bridging the gap between these two environments. We also show that the convergence rate of our estimator to be of the order of n^{-1/4}n^{-1/4}, where nn is the number of training samples. In simulation, we demonstrate how our method effectively approximates and evaluates performance on Gridworld, Cartpole, and Reacher environments on a range of policies. We also show that the our method is able to estimate performance of a 7 DOF robotic arm using the simulator and remotely collected data from the robot in the real world.

In this paper, we propose a novel object-based visual-inertial navigation system fully embedded in a matrix Lie group and built upon the invariant Kalman filtering theory. Specifically, we focus on relative pose measurements of objects and derive an error equation at the associated tangent space. We prove that the observability property does not suffer from the filter inconsistency and nonlinear error terms are identically zero at the object initialization. A thorough Monte-Carlo simulation reveals that our approach yields consistent estimates and is very robust to a large initial state uncertainty. Further-more, we demonstrate a real-world application to the KITTI dataset with a deep neural network-based 3D object detector. Experimental results report that noises on pose measurements follow a Gaussian-like density matching our assumption. The proposed method improves the localization and object global mapping accuracy by probabilistically accounting for inertial readings and object pose uncertainties at multiple views.