IROS 2020 论文列表

This paper aims to develop time-varying virtual constraint controllers that allow stable and agile bounding gaits for full-order hybrid dynamical models of quadrupedal locomotion. As opposed to state-based nonlinear controllers, time-varying controllers can initiate locomotion from zero velocity. Motivated by this property, we investigate the stability guarantees that can be provided by the time-varying approach. In particular, we systematically establish necessary and sufficient conditions that guarantee exponential stability of periodic orbits for time-varying hybrid dynamical systems utilizing the Poincaré return map. Leveraging the results of the presented proof, we develop time-varying virtual constraint controllers to stabilize bounding gaits of a 14 degree of freedom planar quadrupedal robot, Minitaur. A framework for choosing the parameters of virtual constraint controllers to achieve exponential stability is shown, and the feasibility of the analytical results is numerically validated in full-order simulation models of Minitaur.

This paper addressed a challenging problem of wheeled-legged robots with high degrees of freedom exploring in unknown rough environments. The proposed method works as a pipeline to achieve prioritized exploration comprising three primary modules: traversability analysis, frontier-based exploration and hybrid locomotion planning. Traversability analysis provides robots an evaluation about surrounding terrain according to various criteria ( roughness, slope etc.) and other semantic information (small step, stair, bridge etc.), while novel gravity point frontier-based exploration algorithm can effectively decide which direction to go even in unknown environments based on robots' current pose and desired one. Given all these information, hybrid locomotion planner will generate a path with motion mode (driving or walking) encoded by optimizing among different objectives and constraints. Lastly, our approach was well verified in both simulation and experiment on a wheeled quadrupedal robot Pholus.

This paper presents a tangle-free frontier based exploration algorithm for planar mobile robots equipped with limited length and anchored tethers. After planning a path to the closest point in the frontier between free and unknown space, the robot computes an estimate of the future length of its tether and decides, by comparing the anticipated length with the minimum possible tether length, whether the path should be followed or not. If the anticipated tether is longer than the minimum tether by a function of the expected radius of the obstacles, a path planner with homotopic constraints is used to plan a path that brings the robot tether to the same homotopy class of the shortest tether. This behavior will not only limit the tether length but also will prevent tether entangling on the obstacles of the environment. We evaluate our method in different simulated environments and illustrate the approach with an actual tethered robot.

How to explore environments is one of the most critical factors for the performance of an agent in reinforcement learning. Conventional exploration strategies such as ε-greedy algorithm and Gaussian exploration noise simply depend on pure randomness. However, it is required for an agent to consider its training progress and long-term usefulness of actions to efficiently explore complex environments, which remains a major challenge in reinforcement learning. To address this challenge, we propose a novel exploration method that selects actions based on their validity. The key idea behind our method is to estimate the validity of actions by leveraging zero avoiding property of kullback-leibler divergence to comprehensively evaluate actions in terms of both exploration and exploitation. We also introduce a framework that allows an agent to explore efficiently in environments where reward is sparse or cannot be defined intuitively. The framework uses expert demonstrations to guide an agent to visit task-relevant state space by combining our exploration strategy with imitation learning. We demonstrate our exploration strategy on several tasks ranging from classical control tasks to high-dimensional urban autonomous driving scenarios at roundabout. The results show that our exploration strategy encourages an agent to visit task-relevant state space to enhance validity of actions, outperforming several previous methods.

Tendrils are common stable structures in nature and are used for sensing, actuation, and geometrical stiffness modulation. In this paper, for the first time we exploit the helical geometry of a shape memory alloy (SMA) tendril as a simple to fabricate highly dexterous robotic continuum tentacle that we called Active Tendril-Backbone Robot (ATBR). This is achieved via partial (120 deg) activation of single helix turns resulting in backbone directional bendings. A 141.5 mm prototype (130 mm when fully compressed) has been fabricated and a simple theoretical framework is proposed and experimentally validated for modeling of the tentacle configuration. The manipulator has five 2-DOF joints capable of reaching bending angles of up to 54.5 deg and angular speed of up to 6.8 deg/s. The dexterity of the manipulator is showcased empirically in reaching complex configurations and simple navigation through confined space of a curving path.

In this paper, we propose a method to use semantic information to improve the use of map priors in a sparse, feature-based MonoSLAM system. To incorporate the priors, the features in the prior and SLAM maps must be associated with one another. Most existing systems build a map using SLAM and then align it with the prior map. However, this approach assumes that the local map is accurate, and the majority of the features within it can be constrained by the prior. We use the intuition that many prior maps are created to provide semantic information. Therefore, valid associations only exist if the features in the SLAM map arise from the same kind of semantic object as the prior map. Using this intuition, we extend ORB-SLAM2 using an open source pre-trained semantic segmentation network (DeepLabV3+) to incorporate prior information from Open Street Map building footprint data. We show that the amount of drift, before loop closing, is significantly smaller than that for original ORB-SLAM2. Furthermore, we show that when ORB-SLAM2 is used as a prior-aided visual odometry system, the tracking accuracy is equal to or better than the full ORB-SLAM2 system without the need for global mapping or loop closure.

We present the sequential correlation network (SCN) to improve concurrent activity detection. SCN combines a recurrent neural network and a correlation model hierarchically to model the complex correlations and temporal dynamics of concurrent activities. SCN has several advantages that enable effective learning even from a small dataset for real-world deployment. Unlike the majority of approaches assuming that each subject performs one activity at a time, SCN is end-to- end trainable, i.e., it can automatically learn the inclusive or exclusive relations of concurrent activities. SCN is lightweight in design using only a small set of learnable parameters to model the spatio-temporal correlations of activities. This also enhances the explainability of the learned parameters. Furthermore, the learning of SCN can benefit from the initialization using semantically meaningful priors. We evaluate the proposed method against the state-of-the-art method on two benchmark datasets with human skeletal data, SCN achieves comparable performance to the SOTA but with much faster inference speed and less memory usage.

We propose a novel excavation (i.e., digging) trajectory planning framework for industrial autonomous robotic excavators, which emulates the strategies of human expert operators to optimize the excavation of (complex/unmodellable) soils while also upholding robustness and safety in practice. First, we encode the trajectory with dynamic movement primitives (DMP), which is known to robustly preserve qualitative shape of the trajectory and attraction to (variable) end-points (i.e., start-points of swing/dumping), while also being data-efficient due to its structure, thus, suitable for our purpose, where expert data collection is expensive. We further shape this DMPbased trajectory to be expert-emulating, by learning the shaping force of the DMP-dynamics from the real expert excavation data via a neural network (i.e., MLP (multi-layer perceptron)). To cope with (possibly dangerous) underground uncertainties (e.g., pipes, rocks), we also real-time modulate the expert-emulating (nominal) trajectory to prevent excessive build-up of excavation force by using the feedback of its online estimation. The proposed framework is then validated/demonstrated by using an industrial-scale autonomous robotic excavator, with the associated data also presented here.

In this paper an experimental study of set-based task-priority kinematic control for a dual-arm mobile robot is developed. The control strategy for the coordination of the two manipulators and the mobile base relies on the definition of a set of elementary tasks to be properly handled depending on their functional role. In particular, the tasks have been grouped into three categories: safety, operational and optimization tasks. The effectiveness of the resulting task hierarchy has been validated through experiments on a Kinova Movo robot, in a domestic use case scenario.

This work presents the deployment of UAVs for the exploration of clouds, from the system architecture and simulation tests to a real-flight campaign and trajectory analyzes. Thanks to their small size and low altitude, light UAVs have proven to be adapted for in-situ cloud data collection. The short life time of the clouds and limited endurance of the planes require to focus on the area of maximum interest to gather relevant data. Based on previous work on cloud adaptive sampling, the article focuses on the overall system architecture, the improvements made to the system based on preliminary tests and simulations, and finally the results of a field campaign. The Barbados experimental flight campaign confirmed the capacity of the system to map clouds and to collect relevant data in dynamic environment, and highlighted areas for improvement.

The authors have investigated underactuated locomotion robots with an inner wobbling mass, it is discovered that the wobbling mass controls the gait speed by entrainment. Supplying the wobbling from outside, outer wobbling entrains load objects and controls the transferring speed. In this paper, we propose a vibratory conveyor system based on the frequency entrainment of a limit cycle walker. The conveyance plate is vibrated by an active rimless wheel, and the system conveys a passive rimless wheel which is defined as a load object. The vibration entrains transferring of the passive rimless and controls the conveyance speed. First, we introduce the prototype experimental system and its mathematical model. Second, we report basic behavior of the passive rimless with regards to the outer vibration and results of frequency analysis through the numerical simulation. Third, we experimentally verify the results of the numerical simulation. The active rimless wheel entrains the walking frequency of the passive rimless wheel in both the simulations and the experiments.

In this paper we perform an extensive experimental evaluation of three planar target based 3D-LIDAR camera calibration algorithms, on a sensor suite consisting multiple 3D-LIDARs and cameras, assessing their robustness to random initialization and by using metrics like Mean Line Re-projection Error (MLRE) and Factory Stereo Calibration Error. We briefly describe each method and provide insights into practical aspects like ease of data collection. We also show the effect of noisy sensor on the calibration result and conclude with a note on which calibration algorithm should be used under what circumstances.

Autonomous navigation missions require online decision making abilities, in order to choose from a given set of candidate actions an action that will lead to the best outcome. In a partially observable setting, decision making under uncertainty, also known as belief space planning (BSP), involves reasoning about belief evolution considering realizations of future observations. Yet, when candidate actions lead the robot to an unknown environment the decision making mission becomes a very challenging problem since without a map it is hard to foresee future observations. In this paper we develop a data-driven approach for predicting a distribution over an unexplored map, generating future observations, and combining these observations within BSP. We examine our approach and compare it to existing BSP methods in a Gazebo simulation, and demonstrate it often yields improved performance.

We consider a scenario in which an autonomous vehicle equipped with a downward facing camera operates in a 3D environment and is tasked with searching for an unknown number of stationary targets on the 2D floor of the environment. The key challenge is to minimize the search time while ensuring a high detection accuracy. We model the sensing field using a multi-fidelity Gaussian process that systematically describes the sensing information available at different altitudes from the floor. Based on the sensing model, we design a novel algorithm called Expedited Multi-Target Search (EMTS) that (i) addresses the coverage-accuracy trade-off: sampling at locations farther from the floor provides wider field of view but less accurate measurements, (ii) computes an occupancy map of the floor within a prescribed accuracy and quickly eliminates unoccupied regions from the search space, and (iii) travels efficiently to collect the required samples for target detection. We rigorously analyze the algorithm and establish formal guarantees on the target detection accuracy and the detection time. We illustrate the algorithm using a simulated multi-target search scenario.

The musculoskeletal humanoid has various biomimetic benefits, and the redundant muscle arrangement is one of its most important characteristics. This redundancy can achieve fail-safe redundant actuation and variable stiffness control. However, there is a problem that the maximum joint angle velocity is limited by the slowest muscle among the redundant muscles. In this study, we propose two methods that can exceed the limited maximum joint angle velocity, and verify the effectiveness with actual robot experiments.

In gait training for walking rehabilitation of patients with stroke hemiplegia or bone joint conditions such as fractures, it is important to recognize the load of the affected lower limbs for improving gait ability and avoiding risks such as re-fractures. A weight scale is used at the actual rehabilitation site to recognize the load. However, in this situation, the trainee must look down to verify whether the scale and their walking posture are correct. In addition, the trainee generally cannot read the load value accurately. Therefore, we have developed a system that can show the load in real time on an eye-level display. By using this system, we expect the patients to be able to perform gait training smoothly while recognizing the state of walking. In this paper, we have reported the results of a clinical trial held at a rehabilitation hospital and an examination of the screen-indicated methods.

Spiking Neuronal Networks (SNNs) realized in neuromorphic hardware lead to low-power and low-latency neuronal computing architectures. Neuromorphic computing systems are most efficient when all of perception, decision making, and motor control are seamlessly integrated into a single neuronal architecture that can be realized on the neuromorphic hardware. Many neuronal network architectures address the perception tasks, while work on neuronal motor controllers is scarce. Here, we present an improved implementation of a neuromorphic PID controller. The controller was realized on Intel's neuromorphic research chip Loihi and its performance tested on a drone, constrained to rotate on a single axis. The SNN controller is built using neuronal populations, in which a single spike carries information about sensed and control signals. Neuronal arrays perform computation on such sparse representations to calculate the proportional, derivative, and integral terms. The SNN PID controller is compared to a PID controller, implemented in software, and achieves a comparable performance, paving the way to a fully neuromorphic system in which perception, planning, and control are realized in an on-chip SNN.

Although they take many forms, legged robots rely upon springs to achieve high speed, dynamic locomotion. In this paper we examine the effect of adding parallel springs to robots that rely on virtual compliance. Specifically, we consider the trade-off between energetic efficiency and leg versatility that comes while using Parallel Elastic Actuators (PEAs). To do this, we vary the ratio of physical to virtual compliance for legged systems using a) a modified SLIP model, b) a single legged hopping robot, and c) a multibody simulation of the quadruped robot LLAMA. In each case we show that having a small physical compliance significantly improves the efficiency while also maintaining the robot's versatility.

In this research, we propose a method of estimating object class and orientation given multiple input images assuming the relative camera orientations are known. Input images are transformed to descriptors on 2-D manifolds defined for each class of object through a CNN, and the object class and orientation that minimize the distance between input descriptors and the descriptors associated with the estimated object class and orientation are selected. The object orientation is further optimized by interpolating the viewpoints in the database.The usefulness of the proposed method is demonstrated by comparative evaluation with other methods using publicly available datasets. The usefulness of the proposed method is also demonstrated by recognizing images taken from the cameras on our humanoid robot using our own dataset.