ICRA 2018 Papers

The Simultaneous Localization And Mapping (SLAM) problem has been well studied in the robotics community, especially using mono, stereo cameras or depth sensors. 3D depth sensors, such as Velodyne LiDAR, have proved in the last 10 years to be very useful to perceive the environment in autonomous driving, but few methods exist that directly use these 3D data for odometry. We present a new low-drift SLAM algorithm based only on 3D LiDAR data. Our method relies on a scan-to-model matching framework. We first have a specific sampling strategy based on the LiDAR scans. We then define our model as the previous localized LiDAR sweeps and use the Implicit Moving Least Squares (IMLS) surface representation. We show experiments with the Velodyne HDL32 with only 0.40% drift over a 4 km acquisition without any loop closure (i.e., 16 m drift after 4 km). We tested our solution on the KITTI benchmark with a Velodyne HDL64 and ranked among the best methods (against mono, stereo and LiDAR methods) with a global drift of only 0.69%.

In the context of learning from demonstration, human examples are usually imitated in either Cartesian or joint space. However, this treatment might result in undesired movement trajectories in either space. This is particularly important for motion skills such as striking, which typically imposes motion constraints in both spaces. In order to address this issue, we consider a probabilistic formulation of dynamic movement primitives, and apply it to adapt trajectories in Cartesian and joint spaces simultaneously. The probabilistic treatment allows the robot to capture the variability of multiple demonstrations and facilitates the mixture of trajectory constraints from both spaces. In addition to this proposed hybrid space learning, the robot often needs to consider additional constraints such as motion smoothness and joint limits. On the basis of Jacobian-based inverse kinematics, we propose to exploit robot null-space so as to unify trajectory constraints from Cartesian and joint spaces while satisfying additional constraints. Evaluations of hand-shaking and striking tasks carried out with a humanoid robot demonstrate the applicability of our approach.

This paper considers the motion control and task planning problem of mobile robots under complex high-level tasks and human initiatives. The assigned task is specified as Linear Temporal Logic (LTL) formulas that consist of hard and soft constraints. The human initiative influences the robot autonomy in two explicit ways: with additive terms in the continuous controller and with contingent task assignments. We propose an online coordination scheme that encapsulates (i) a mixed-initiative continuous controller that ensures all-time safety despite of possible human errors, (ii) a plan adaptation scheme that accommodates new features discovered in the workspace and short-term tasks assigned by the operator during run time, and (iii) an iterative inverse reinforcement learning (IRL) algorithm that allows the robot to asymptotically learn the human preference on the parameters during the plan synthesis. The results are demonstrated by both realistic human-in-the-loop simulations and experiments.

Existing control systems for optical manipulation of multiple micro-objects do not allow users to interact with the systems during manipulation to deal with unexpected events, while ensuring stability of the overall control systems. In this paper, we propose a robotic control technique for human-guided optical manipulation of multiple micro-objects using robotic tweezers. Humans, when necessary, are able to interact or intervene with an automated optical manipulation system in a stable manner, and thus guiding a group of micro-objects to be manipulated towards a desired region while ensuring collision avoidance during manipulation. Both the ability of humans, in term of decision making, when needed, and the advantages of an automated optical manipulation system which provides precise and productive manipulation of micro-objects, are consolidated into one single technique. A theoretical foundation is developed and investigated to achieve the control objective. Experimental results are presented to illustrate the effectiveness of the proposed control technique.

Dexterous manipulation with robotic hands can be achieved using object-level impedance control strategies, which allow intuitive regulation of object position, external environmental interactions, and grasp forces. However, for grasp stability, object stiffness gains are limited by the inherent compliance of the robotic system, object size/shape, and applied grasp forces, which can lead to restricted manipulation capabilities. In this work, we first use analytical modeling techniques to explore the theoretical passivity bounds on object stiffness control gains to ensure grasp stability. Then, an object-space stiffness control algorithm is developed for the Anatomically Correct Testbed (ACT) hand, a robotic hand designed to replicate the complex tendon and joint structure of the human hand, and grasp stability bounds are experimentally tested for various task scenarios. Finally, inspired by the hierarchical structure of the human neuromuscular system, we develop a novel control strategy that implements low-level stiffness in muscle-space, while also emulating a separately defined object-space stiffness in quasi-static conditions. Experimental results demonstrate that this control strategy increases achievable object stiffness without sacrificing grasp stability, leading to significantly increased manipulation capabilities.

This paper presents a practical method to identify body segments inertial parameters of a human-exoskeleton system using affordable and easy-to-use sensors. First, the joints and the base kinematics are estimated based on the use of an extended Kalman filter and QR visual markers. Then, joints kinematics are used in a dynamic identification pipeline together with the ground reaction force and moments collected with an affordable Wii Balance Board. The identification process is done using an augmented regressor matrix to identify at once each segment mass, center of mass 3D position and inertia tensor elements of both human locomotor apparatus and exoskeleton. The proposed method is able to accurately estimate external force and moments, with less than 6 % of normalized RMS difference in average, and is experimentally validated with a subject wearing a full lower limb exoskeleton.

The state features available to a robot define the variables on which the learning computation depends. However, little prior work considers feature selection in the context of deploying a general-purpose robot able to learn new tasks. In this work, we explore human-driven feature selection in which a robotic agent can identify useful features with the aid of a human user, by extracting information from users about which features are most informative for discriminating between classes of objects needed for a given task (e.g. sorting groceries). The research questions examine (a) whether a domain expert is able to identify a subset of informative task features, (b) whether human selected features will enable the agent to classify unseen examples as accurately as using computational feature selection, and (c) if the interaction strategy used to elicit the information from the user impacts the quality of resulting feature selection. Toward that end, we conducted a user study with 30 participants on campus, given a multi-class classification task and one of five different approaches for conveying information about informative features to a robot learner. Our findings show that when features are semantically interpretable, human feature selection is effective in LfD scenarios because it is able to outperform computational methods when there is limited training data, yet still remains on-par with computational methods as the training sample size increases.

Shared control and shared autonomy play an important role in assistive technologies, allowing the offloading of the cognitive burden required for control from the user to the intelligent robotic device. In this context, electrophysiological measures of error detection, directly measured from a person's brain activity as Error-related Potentials (ErrPs), can be exploited to provide passive adaptation of an external semi-autonomous system to the human. This concept was implemented in an online robot learning task, where user's evaluation of the robot's actions, in terms of detected ErrP, was exploited to update a reward function in a Reinforcement Learning (RL) framework. Results from both simulated and experimental studies show that the introduction of human evaluation in the robot learning loop allows for: (1) the acceleration of optimal policy learning in a target reaching task, (2) the introduction of a further degree of control in robot learning, namely identification of one among multiple targets, according to the user's will. Overall, presented results support the potential of human-robot co-adaptive and co-operative strategies to develop human-centered assistive technologies.

Current motion capture (MoCap) systems generally require markers and multiple calibrated cameras, which can be used only in constrained environments. In this work we introduce a drone-based system for 3D human MoCap. The system only needs an autonomously flying drone with an on-board RGB camera and is usable in various indoor and outdoor environments. A reconstruction algorithm is developed to recover full-body motion from the video recorded by the drone. We argue that, besides the capability of tracking a moving subject, a flying drone also provides fast varying viewpoints, which is beneficial for motion reconstruction. We evaluate the accuracy of the proposed system using our new DroCap dataset and also demonstrate its applicability for MoCap in the wild using a consumer drone.

The coordination problems arising in a team of autonomous mobile robots have received a lot of attention in the distributed robotics community. Along those lines, we study in this paper the problem of coordinating autonomous mobile robots to reposition on a convex hull so that each robot sees all others. In particular, we consider non-transparent fat robots operating in the 2-dimensional plane. They are abstracted as unit discs and they make local decisions with vision being the only mean of coordination among them. We develop a (deterministic) distributed algorithm that solves the problem for a team of N ≥ 3 fat robots in O(N) time avoiding collisions under the semi-synchronous scheduler. The main idea is to enforce the robots to reach a configuration in which (i) the robots' centers form a convex hull; (ii) all robots are on the convex hull's boundary; and (iii) each robot can see all other robots. The result is achieved assuming some reasonable conditions on the input configuration and showing that starting from any input configuration that satisfies our conditions, robots reach such a configuration in linear time and terminate.

Complete coverage, which is the foundation of many robotic applications, aims to cover an area as quickly as possible. This study investigates the time-limited version of multi-robot complete coverage problem, that is, to find the least number of robots and allocate tasks properly to them such that they can finish a known mission within the time limit. This version of problem can be tackled straightforwardly based on optimizing the task-allocation to a fixed number of robots and enumerating the number. However, the number-fixed problem is NP-hard and the existing algorithm for the number-fixed problem allows intersecting tasks (possibly causing robots' interference) and endures high approximation factor. In this study, the time-limited complete coverage problem is tackled with a multi-objective approach, instead of enumerating robots' number and optimizing each number-fixed problem one by one. The multi-objective GA, Mofint, at first estimates the lower and upper bounds of the number of robots. It abstracts each task as a weighted node of a graph. Then, Mofint evolves individuals, each individual being a forest containing a certain number (within the bounds) of non-intersecting trees. Mofint can finally obtain higher precision than existing work with less time: the approximation factor for Mofint is 1.1 to 1.5 times the ideal allocation when robots' number is fixed, while for existing work is 1.5 to 2. Due to its higher precision, the least number of robots obtained in the experiments by Mofint is 0.6 times of existing work.

This work presents a method for automated generation of navigation plan for autonomous or robotic surface vessel. Historical Automatic Identification System (AIS) data is of significant value to this problem. The method joins AIS locations of a same vessel at different time and locations in a region into a route. Next, it automatically computes navigation plans using nearest neighbour based path retrieval relying on two representations, Ship Feature and Navigation Feature. Before starting service, existing AIS records in the form of ship properties and corresponding route are preprocessed and stored in the form of Ship and Navigation Feature. During online retrieval, given input constraints in vector form, nearest neighbour of this query vector in the same space is found and corresponding path of the neighbour is returned as recommended path. Analysis was done in four and two dimensional spaces for Ship and Navigation Feature respectively. Application of the method is demonstrated in two regions of Australian, covering Bass Strait and Great Australian Bight.

In recent years, vision-based analysis systems of micro-objects in a microchannel have been actively developed. However, it is difficult to focus on high-speed micro-objects in a microchannel because the general height of a microchannel is approximately 10-100 μm, whereas the depth of focus of the objective lens is approximately 1-4 μm. Therefore, we propose a high-speed well-focused image-capturing microscope, which is a system with an objective lens attached to a vibration machine that moves the focus position rapidly by oscillating it up and down to capture well-focused images using a histogram-based algorithm. The proposed microscope system is verified experimentally to capture well-focused images of moving micro- objects.

We present a deep convolutional neural network (CNN) architecture for high-precision depth estimation by jointly utilizing sparse 3D LiDAR and dense stereo depth information. In this network, the complementary characteristics of sparse 3D LiDAR and dense stereo depth are simultaneously encoded in a boosting manner. Tailored to the LiDAR and stereo fusion problem, the proposed network differs from previous CNNs in the incorporation of a compact convolution module, which can be deployed with the constraints of mobile devices. As training data for the LiDAR and stereo fusion is rather limited, we introduce a simple yet effective approach for reproducing the raw KITTI dataset. The raw LiDAR scans are augmented by adapting an off-the-shelf stereo algorithm and a confidence measure. We evaluate the proposed network on the KITTI benchmark and data collected by our multi-sensor acquisition system. Experiments demonstrate that the proposed network generalizes across datasets and is significantly more accurate than various baseline approaches.

In this paper, we propose a probabilistic MLN-based model for spatial context disambiguation. This model serves as a solution for the problem of incomplete knowledge in High-level task planning. By applying the state of the art MLN probabilistic reasoning such as MCSAT, we determine the concept class of the current spatial context of the robot and contribute by combining semantic spatial relations with observed data at different timesteps. The inherent uncertainty of robot dynamic environments makes the proposed approach suitable to deal with partial observability and sensing limitations of robots. Simulation experiments and evaluation results are presented to validate our model.

The present paper investigates if and in which conditions does the implementation of high stiffness controllers increase the performances of human dyads during comanipulative tasks in physical Human-Human Interaction (pHHI) settings. Two experiments are conducted which cover two fundamental aspects of pHHI: low-level interactions allowing interpersonal coordination, and high-level interactions allowing common decision-making and negotiation of strategies. The results of these experiments show that high stiffness is not necessary for good performances when the task only requires low-level interactions. On the contrary, when dealing with high-level interactions, higher stiffness increases task performance. The results presented highlight the importance of the quality of teleoperated control in setups used for the study of pHHI.

In this paper, we propose novel methods suitable for online real time whole body master-slave control with real life-sized humanoid robot. We conducted some dynamic whole body master-slave experiment with life-sized humanoid robot, and we achieved speedier and flexible master-slave operation compared to conventional study. Conventionally, master-slave operations with humanoid robots were available with only the upper body of the humanoid robot, and the COM movement was limited to be static. In our previous study, we introduced LIP model based restrictions to ensure the balance stability. In this study, we extend the safety restrictions by introducing foot landing delay prediction and trajectory smoothing method suitable for real robot. We conducted master-slave tennis swing experiment and high kick motion experiment with life-sized humanoid robot “JAXON”, and we evaluated the effectiveness of our proposed methods and system.

This paper describes the design of a high dynamic range (HDR) six-axis force/torque sensor. The sensor is composed of a high-rigidity flexure element detecting large force and a low-rigidity flexure element detecting small force. The overload on the low-rigidity flexure element is prevented by an overload protection mechanism. An HDR measurement is achieved by combining the outputs of the two flexure elements. A loading test for the designed sensor is performed, and the results indicate that the six-axis sensor measures force with a dynamic range from 0.01N to 1000 N.

Physical sampling of water for off-site analysis is necessary for many applications like monitoring the quality of drinking water in reservoirs, understanding marine ecosystems, and measuring contamination levels in fresh-water systems. In this paper, the focus is on algorithms for efficient measurement and sampling using a multi-robot, data-driven, water-sampling behavior, where autonomous surface vehicles plan and execute water sampling using the chlorophyll density as a cue for plankton-rich water samples. We use two Autonomous Surface Vehicles (ASVs), one equipped with a water quality sensor and the other equipped with a water-sampling apparatus. The ASV with the sensor acts as an explorer, measuring and building a spatial map of chlorophyll density in the given region of interest. The ASV equipped with the water sampling apparatus makes decisions in real time on where to sample the water based on the suggestions made by the explorer robot. We evaluate the system in the context of measuring chlorophyll distributions. We do this both in simulation based on real geophysical data from MODIS measurements, and on real robots in a water reservoir. We demonstrate the effectiveness of the proposed approach in several ways including in terms of mean error in the interpolated data as a function of distance traveled.