IROS 2020 论文列表

This paper presents VCU-RVI, a new visual inertial odometry (VIO) benchmark with a set of diverse data sequences in different indoor scenarios. The benchmark was captured using an Structure Core (SC) sensor, consisting of an RGB-D camera and an IMU. It provides aligned color and depth images with 640×480 resolution at 30 Hz. The camera's data is synchronized with the IMU's data at 100 Hz. Thirty-nine data sequences covering a total of ~3.7 kilometers trajectory were recorded in various indoor environments by two experimental setups: hand-holding the SC sensor or installing it on a wheeled robot. For the data sequences from the handheld SC, some were recorded in our laboratory under three challenging conditions: fast sensor motion, radical illumination changing, and dynamic objects, and the rest were collected in various indoor spaces outside the laboratory in the East Engineering Building, including corridors, halls, and stairways, during long-distance navigation scenarios. For the data sequences captured using the wheeled robot, half of them were recorded with sufficient IMU excitation in the beginning of the sequence, to meet the need of testing the VIO methods with the requirement of sufficient motion conditions for initialization. We placed three bumpers on the floor of the lab to create an uneven terrain to make the robot motion 6-DOF. The sequences also include data collected from navigational courses with a long trajectory. For trajectory evaluation, a motion capture system is used to generate accurate pose data (at a rate of 120 Hz), which will be used as the ground truth. We conducted experiments to evaluate the state-of-the-art VIO algorithms using our benchmark. These algorithms together with the evaluation tools and the VCU-RVI dataset are made publicly available.

The SPIR, Submersible Pylon Inspection Robot, is developed to provide an innovative and practical solution to keep workers safe during maintenance of underwater structures in shallow waters, which involves working in dangerous water currents, and high-pressure water-jet cleaning. More advanced than work-class Remotely Operated Vehicles technology, the SPIR is automated and required minimum involvement of humans into the working process, thus effectively lowered the learning curve required to conduct work. To make SPIR operate effectively in poor visibility and highly disturbed environments, the multiple new technologies are developed and implemented into the system, including SBL-SONAR-based navigation, 6-DOF stabilisation, and vision-based 3D mapping. Extensive testing and field trials in various bridges are conducted to verify the robotic system. The results demonstrate the suitability of the SPIR in substituting humans for underwater hazardous tasks such as autonomous cleaning and inspection of bridge and wharf piles.

To act effectively in its environment, a cognitive robot needs to understand the causal dependencies of all intermediate actions leading up to its goal. For example, the system has to infer that it is instrumental to open a cupboard door before trying to grasp an object inside the cupboard. In this paper, we introduce a novel learning method for extracting instrumental dependencies by following the scientific approach of observations, generation of causal hypotheses, and testing through experiments. Our method uses a virtual reality dataset containing observations from human activities to generate hypotheses about causal dependencies between actions. It detects pairs of actions with a high temporal co-occurrence and verifies if one action is instrumental in executing the other action through mental simulation in a virtual reality environment which represents the system's mental model. Our system is able to extract all present instrumental action dependencies while significantly reducing the search space for mental simulation, resulting in a 6-fold reduction in computational time.

We present the Pluggable Distributed Resource Allocator (PDRA), a middleware for distributed computing in heterogeneous mobile robotic networks. PDRA enables autonomous robotic agents to share computational resources for computationally expensive tasks such as localization and path planning. It sits between an existing single-agent planner/executor and existing computational resources (e.g. ROS packages), intercepts the executor's requests and, if needed, transparently routes them to other robots for execution. PDRA is pluggable: it can be integrated in an existing single-robot autonomy stack with minimal modifications. Task allocation decisions are performed by a mixed-integer programming algorithm, solved in a shared-world fashion, that models CPU resources, latency requirements, and multi-hop, periodic, bandwidth-limited network communications; the algorithm can minimize overall energy usage or maximize the reward for completing optional tasks. Simulation results show that PDRA can reduce energy and CPU usage by over 50% in representative multi-robot scenarios compared to a naive scheduler; runs on embedded platforms; and performs well in delay- and disruption-tolerant networks (DTNs). PDRA is available to the community under an open-source license.

Snake robots have the potential to locomote through tightly packed spaces, but turning effectively within unmodelled and unsensed environments remains challenging. Inspired by a behavior observed in the tiny nematode worm C. elegans, we propose a novel in-place turning gait for elongated limbless robots. To simplify the control of the robots' many internal degrees-of-freedom, we introduce a biologically-inspired template in which two co-planar traveling waves are superposed to produce an in-plane turning motion, the omega turn. The omega turn gait arises from modulating the wavelengths and amplitudes of the two traveling waves. We experimentally test the omega turn on a snake robot, and show that this turning gait outperforms previous turning gaits: it results in a larger angular displacement and a smaller area swept by the body over a gait cycle, allowing the robot to turn in highly confined spaces.

In this paper, we present a large dataset with a variety of mobile mapping sensors collected using a handheld device carried at typical walking speeds for nearly 2.2 km around New College, Oxford as well as a series of supplementary datasets with much more aggressive motion and lighting contrast. The datasets include data from two commercially available devices - a stereoscopic-inertial camera and a multi-beam 3D LiDAR, which also provides inertial measurements. Additionally, we used a tripod-mounted survey grade LiDAR scanner to capture a detailed millimeter-accurate 3D map of the test location (containing ~290 million points). Using the map, we generated a 6 Degrees of Freedom (DoF) ground truth pose for each LiDAR scan (with approximately 3 cm accuracy) to enable better benchmarking of LiDAR and vision localisation, mapping and reconstruction systems. This ground truth is the particular novel contribution of this dataset and we believe that it will enable systematic evaluation which many similar datasets have lacked. The large dataset combines both built environments, open spaces and vegetated areas so as to test localisation and mapping systems such as vision-based navigation, visual and LiDAR SLAM, 3D LiDAR reconstruction and appearance-based place recognition, while the supplementary datasets contain very dynamic motions to introduce more challenges for visual-inertial odometry systems. The datasets are available at:ori.ox.ac.uk/datasets/newer-college-dataset.

This paper presents the design of the Multi-material Actuator for Variable Stiffness (MAVS), which consists of an inflatable soft fabric actuator fixed between two layers of rigid retainer pieces. The MAVS is designed to be integrated with a soft robotic ankle-foot orthosis (SR-AFO) exosuit to aid in supporting the human ankle in the inversion/eversion directions. This design aims to assist individuals affected with chronic ankle instability (CAI) or other impairments to the ankle joint. The MAVS design is made from compliant fabric materials, layered and constrained by thin rigid retainers to prevent volume increase during actuation. The design was optimized to provide the greatest stiffness and least deflection for a beam positioned as a cantilever with a point load. Geometric programming of materials was used to maximize stiffness when inflated and minimize stiffness when passive. An analytic model of the MAVS was created to evaluate the effects in stiffness observed by varying the ratio in length between the rigid pieces and the soft actuator. A finite element analysis (FEA) was generated to analyze and predict the behavior of the MAVS prior to fabrication. The results from the analytic model and FEA study were compared to experimentally obtained results of the MAVS. The MAVS with the greatest stiffness was observed when the gap between the rigid retainers was smallest and the rigid retainer length was smallest. The MAVS design with the highest stiffness at 100 kPa was determined, which required 26.71 ± 0.06 N to deflect the actuator 20 mm, and a resulting stiffness of 1,335.5 N/m and 9.1% margin of error from the model predictions.

In this paper, we propose a flexible mapping scheme that uses a masking function (mask) to focus the attention of a pose graph SLAM (Simultaneous Localization and Mapping) system. The masking function takes the robot's observations and returns true if the robot is in an important location. State-of-the-art methods in SLAM generate dense metric lidar maps, creating precise maps at a high computational cost by storing lidar scans for each pose node and continually attempting to close loops. In many cases, trying to always make loop closures is unnecessary for localization and even risky because of perceptual aliasing and false positives. By masking out these less useful positions, our method can create more accurate maps despite performing far fewer scan matches. We evaluate our system with three simple mask functions on a 2.5 km trajectory with significant angular drift. We compare the number of scan matches performed under each mask as well as the accuracy of the loop closures.

Developments in mobile robot navigation have enabled robots to operate in warehouses, retail stores, and on sidewalks around pedestrians. Various navigation solutions have been proposed, though few as widely adopted as ROS (Robot Operating System) Navigation. 10 years on, it is still one of the most popular navigation solutions1. Yet, ROS Navigation has failed to keep up with modern trends. We propose the new navigation solution, Navigation2, which builds on the successful legacy of ROS Navigation. Navigation2 uses a behavior tree for navigator task orchestration and employs new methods designed for dynamic environments applicable to a wider variety of modern sensors. It is built on top of ROS2, a secure message passing framework suitable for safety critical applications and program lifecycle management. We present experiments in a campus setting utilizing Navigation2 to operate safely alongside students over a marathon as an extension of the experiment proposed in Eppstein et al. [1]. The Navigation2 system is freely available at https://github.com/ros-planning/navigation2 with a rich community and instructions.

Roads have well defined geometries, topologies, and traffic rules. While this has been widely exploited in motion planning methods to produce maneuvers that obey the law, little work has been devoted to utilize these priors in perception and motion forecasting methods. In this paper we propose to incorporate these structured priors as a loss function. In contrast to imposing hard constraints, this approach allows the model to handle non-compliant maneuvers when those happen in the real world. Safe motion planning is the end goal, and thus a probabilistic characterization of the possible future developments of the scene is key to choose the plan with the lowest expected cost. Towards this goal, we design a framework that leverages REINFORCE to incorporate non-differentiable priors over sample trajectories from a probabilistic model, thus optimizing the whole distribution. We demonstrate the effectiveness of our approach on real-world self-driving datasets containing complex road topologies and multi-agent interactions. Our motion forecasts not only exhibit better precision and map understanding, but most importantly result in safer motion plans taken by our self-driving vehicle. We emphasize that despite the importance of this evaluation, it has been often overlooked by previous perception and motion forecasting works.

In a wide range of applications involving multiple mobile autonomous systems, maneuvering the robots, vehicles or vessels in some sort of formation is a vital component for the overall task performance. Maintaining a specific distance between the platforms or even a relative geometry may greatly enhance sensor performance, provide collision safety, ensure stable vehicle-to-vehicle communication, and is of critical importance when the systems are in some way physically connected. In this paper, we present a flexible leader-follower type formation control algorithm for autonomous robots which is very simple, generic, yet decent in performance. The method applies to any relative geometry between a leader and one or more followers. In addition to testing the algorithm in simulations for a wide range of scenarios, we have performed experiments involving several different autonomous systems, including small Unmanned Aerial Vehicles (UAVs), Autonomous Underwater Vehicles (AUVs) and Unmanned Surface Vehicles (USVs). This includes pairs of USVs physically interconnected by a tow.

This paper details the development, modeling and performance of AmphiSTAR, a novel high-speed amphibious robot. The palm size AmphiSTAR, which belongs to the family of STAR robots, is a "wheeled" robot fitted with propellers at its bottom that allow it to crawl on the ground and run (i.e. hover) on water at high speeds. The AmphiSTAR is inspired by two members of the animal kingdom. It possesses a sprawling mechanism inspired by cockroaches, and it is designed to run on water at high speeds like the Basilisk lizard. We start by presenting the mechanical design of the robot and its control system. Then we model AmphiSTAR when crawling, swimming and running on water. We then report experiments on the robot to measure its lift and thrust forces in its on-water running mode and evaluate its energy consumption. The results show that in the on-water running mode, the lift forces are a function of the work volume of the propellers whereas the thrust forces are a linear function of the propellers' rotating speed. Based on these results, the final version of the 3D printed robot was built and experimentally tested in multiple scenarios. The experimental robot can crawl over the ground with performances similar to the original STAR robot and can attain speeds of 3.6 m/s. The robot can run continuously on water surfaces at speeds of 1.5 m/s. It can also swim (i.e. float while advancing by rotating its propellers) at low speeds and transition from swimming to crawling (see video).

This article presents a planning and control pipeline for legged-wheeled (hybrid) machines. It consists of a Trajectory Optimization based planner that computes references for end-effectors and joints. The references are tracked using a whole-body controller based on a hierarchical optimization approach. Our controller is capable of performing terrain adaptive whole-body control. Furthermore, it computes both torque and position/velocity references, depending on the actuator capabilities. We perform experiments on a Menzi Muck M545, a full size 31 Degrees of Freedom (DoF) walking excavator with five limbs: four wheeled legs and an arm. We show motions that require full-body coordination executed in realistic conditions. To the best of our knowledge, this is the first work that shows the execution of whole-body motions on a full size walking excavator, using all DoFs for locomotion.

We explore using reinforcement learning on single and multi-agent systems such that after learning is finished we can apply a policy zero-shot to new environment sizes, as well as different number of agents and entities. Building off previous work, we show how to map back and forth between the state and action space of a standard Markov Decision Process (MDP) and multi-dimensional tensors such that zero-shot transfer in these cases is possible. Like in previous work, we use a special network architecture designed to work well with the tensor representation, known as the Fully Convolutional Q-Network (FCQN). We show simulation results that this tensor state and action space combined with the FCQN architecture can learn faster than traditional representations in our environments. We also show that the performance of a transferred policy is comparable to the performance of policy trained from scratch in the modified environment sizes and with modified number of agents and entities. We also show that the zero- shot transfer performance across team sizes and environment sizes remains comparable to the performance of training from scratch specific policies in the transferred environments. Finally, we demonstrate that our simulation trained policies can be applied to real robots and real sensor data with comparable performance to our simulation results. Using such policies we can run variable sized teams of robots in a variable sized operating environment with no changes to the policy and no additional learning necessary.

We present a novel paradigm and algorithm for optimal design of underactuated robot platforms in highly-constrained nonconvex parameter spaces. We apply this algorithm to two variants of the mature RoboBee platform, numerically demonstrating predicted performance improvements of over 10% in some cases by algorithmically reasoning about variable effective-mechanical-advantage (EMA) transmissions, higher aspect ratio (AR) wing designs, and force-power tradeoffs. The algorithm can currently be applied to any underactuated mechanical system with one actuated degree of freedom (DOF), and can be easily extended to arbitrary configuration spaces and dynamics.

For many applications, robots will need to be incrementally trained to recognize the specific objects needed for an application. This paper presents a practical system for incrementally training a robot to recognize different object categories using only a small set of visual examples provided by a human. The paper uses a recently developed state-of-the-art method for few-shot incremental learning of objects. After learning the object classes incrementally, the robot performs a table cleaning task organizing objects into categories specified by the human. We also demonstrate the system’s ability to learn arrangements of objects and predict missing or incorrectly placed objects. Experimental evaluations demonstrate that our approach achieves nearly the same performance as a system trained with all examples at one time (batch training), which constitutes a theoretical upper bound.

Chopsticks constitute a simple yet versatile tool that humans have used for thousands of years to perform a variety of challenging tasks ranging from food manipulation to surgery. Applying such a simple tool in a diverse repertoire of scenarios requires significant adaptability. Towards developing autonomous manipulators with comparable adaptability to humans, we study chopsticks-based manipulation to gain insights into human manipulation strategies. We conduct a within-subjects user study with 25 participants, evaluating three different data-collection methods: normal chopsticks, motion-captured chopsticks, and a novel chopstick telemanipulation interface. We analyze factors governing human performance across a variety of challenging chopstick-based grasping tasks. Although participants rated teleoperation as the least comfortable and most difficult-to-use method, teleoperation enabled users to achieve the highest success rates on three out of five objects considered. Further, we notice that subjects quickly learned and adapted to the teleoperation interface. Finally, while motion-captured chopsticks could provide a better reflection of how humans use chopsticks, the teleoperation interface can produce quality on-hardware demonstrations from which the robot can directly learn.

Autonomous vehicles need to plan at the task level to compute a sequence of symbolic actions, such as merging left and turning right, to fulfill people's service requests, where efficiency is the main concern. At the same time, the vehicles must compute continuous trajectories to perform actions at the motion level, where safety is the most important. Task-motion planning in autonomous driving faces the problem of maximizing task-level efficiency while ensuring motion-level safety. To this end, we develop algorithm Task-Motion Planning for Urban Driving (TMPUD) that, for the first time, enables the task and motion planners to communicate about the safety level of driving behaviors. TMPUD has been evaluated using a realistic urban driving simulation platform. Results suggest that TMPUD performs significantly better than competitive baselines from the literature in efficiency, while ensuring the safety of driving behaviors.

In this paper, we propose Belief Behavior Trees (BBTs), an extension to Behavior Trees (BTs) that allows to automatically create a policy that controls a robot in partially observable environments. We extend the semantic of BTs to account for the uncertainty that affects both the conditions and action nodes of the BT. The tree gets synthesized following a planning strategy for BTs proposed recently: from a set of goal conditions we iteratively select a goal and find the action, or in general the subtree, that satisfies it. Such action may have preconditions that do not hold. For those preconditions, we find an action or subtree in the same fashion. We extend this approach by including, in the planner, actions that have the purpose to reduce the uncertainty that affects the value of a condition node in the BT (for example, turning on the lights to have better lighting conditions). We demonstrate that BBTs allows task planning with non-deterministic outcomes for actions. We provide experimental validation of our approach in a real robotic scenario and - for sake of reproducibility - in a simulated one.