IROS 2021 Papers

Proactive inspection is essential for prediction and prevention of rolling stock component failures. The conventional process for inspecting bogies under trains presents significant challenges for inspectors who need to visually check the tight and cluttered environment. We propose a miniature multi-link climbing robot, called BogieBot, that can be deployed inside the undercarriage areas of trains and other large vehicles for inspection and maintenance purposes. BogieBot can carry a visual sensor or manipulator on its main body. The novel compact design utilises six identical couple joints and two mechanically switchable magnetic grippers that together, empower multi-modal climbing and manipulation. The proposed mechanism is kinematically redundant, allowing the robot to perform self-motions in a tight space and manoeuvre around obstacles. The mechanism design and various analyses on the forward and inverse kinematic, work-space, and self-motions of BogieBot are presented. The robot is demonstrated to perform challenging navigation tasks in different scenarios involving simulated complex environments.

A neural network based flexible object manipulation system for a humanoid robot on FPGA is proposed. Although the manipulations of flexible objects using robots attract ever increasing attention since these tasks are the basic and essential activities in our daily life, it has been put into practice only recently with the help of deep neural networks. However such systems have relied on GPU accelerators, which cannot be implemented into the space limited robotic body. Although field programmable gate arrays (FPGAs) are known to be energy efficient and suitable for embedded systems, the model size should be drastically reduced since FPGAs have limited on-chip memory. To this end, we propose "partially" binarized deep convolutional auto-encoder technique, where only an encoder part is binarized to compress model size without degrading the inference accuracy. The model implemented on Xilinx ZCU102 achieves 41.1 frames per second with a power consumption of 3.1 W, which corresponds to 10× and 3.7× improvements from the systems implemented on Core i7 6700K and RTX 2080 Ti, respectively.

We propose a novel benchmark environment for Safe Reinforcement Learning focusing on aquatic navigation. Aquatic navigation is an extremely challenging task due to the non-stationary environment and the uncertainties of the robotic platform, hence it is crucial to consider the safety aspect of the problem, by analyzing the behavior of the trained network to avoid dangerous situations (e.g., collisions). To this end, we consider a value-based and policy-gradient Deep Reinforcement Learning (DRL) and we propose a crossover-based strategy that combines gradient-based and gradient-free DRL to improve sample-efficiency. Moreover, we propose a verification strategy based on interval analysis that checks the behavior of the trained models over a set of desired properties. Our results show that the crossover-based training outperforms prior DRL approaches, while our verification allows us to quantify the number of configurations that violate the behaviors that are described by the properties. Crucially, this will serve as a benchmark for future research in this domain of applications.

In recent years, many learning based approaches have been studied to realize robotic manipulation and assembly tasks, often including vision and force/tactile feedback. How-ever, it is unclear what the baseline state-of-the-art performance is and what the bottleneck problems are. In this work, we evaluate off-the-shelf (OTS) industrial solutions on a recently introduced benchmark, the National Institute of Standards and Technology (NIST) Assembly Task Board. A set of assembly tasks is introduced and baseline methods are provided to understand their intrinsic difficulty. Multiple sensor-based robotic solutions are then evaluated, including hybrid force/motion control and 2D/3D pattern matching. An end-to-end integrated solution that accomplishes the tasks is also provided.The results and findings throughout the study reveal a few noticeable factors that impede the adoptions of the OTS solutions: dependency on expertise, limited applicability, lack of interoperability, no scene awareness or error recovery mechanisms, and high cost. This paper also provides a first attempt of an objective benchmark performance on the NIST Assembly Task Boards as a reference comparison for future works on this problem.

Recent advances in robot learning have enabled robots to become increasingly better at mastering a predefined set of tasks. On the other hand, as humans, we have the ability to learn a growing set of tasks over our lifetime. Continual robot learning is an emerging research direction with the goal of endowing robots with this ability. In order to learn new tasks over time, the robot first needs to infer the task at hand. Task inference, however, has received little attention in the multi-task learning literature. In this paper, we propose a novel approach to continual learning of robotic control tasks. Our approach performs unsupervised learning of behavior embeddings by incrementally self-organizing demonstrated behaviors. Task inference is made by finding the nearest behavior embedding to a demonstrated behavior, which is used together with the environment state as input to a multi-task policy trained with reinforcement learning to optimize performance over tasks. Unlike previous approaches, our approach makes no assumptions about task distribution and requires no task exploration to infer tasks. We evaluate our approach in experiments with concurrently and sequentially presented tasks and show that it outperforms other multi-task learning approaches in terms of generalization performance and convergence speed, particularly in the continual learning setting.

Informed and robust decision making in the face of uncertainty is critical for robots operating in unstructured environments. We formulate this as Bayesian Reinforcement Learning over latent Markov Decision Processes (MDPs). While Bayes-optimality is theoretically the gold standard, existing algorithms scale poorly to continuous state and action spaces. We build on the following insight: in the absence of uncertainty, each latent MDP is easier to solve. We first obtain an ensemble of experts, one for each latent MDP, and fuse their advice to compute a baseline policy. Next, we train a Bayesian residual policy to improve upon the ensemble’s recommendation and learn to reduce uncertainty. Our algorithm, Bayesian Residual Policy Optimization (BRPO), imports the scalability of policy gradient methods and task-specific expert skills. BRPO significantly improves the ensemble of experts and drastically outperforms existing adaptive RL methods, both in simulated and physical robot experiments.

Reinforcement learning methods can achieve significant performance but require a large amount of training data collected on the same robotic platform. A policy trained with expensive data is rendered useless after making even a minor change to the robot hardware. In this paper, we address the challenging problem of adapting a policy, trained to perform a task, to a novel robotic hardware platform given only few demonstrations of robot motion trajectories on the target robot. We formulate it as a few-shot meta-learning problem where the goal is to find a meta-model that captures the common structure shared across different robotic platforms such that data-efficient adaptation can be performed. We achieve such adaptation by introducing a learning framework consisting of a probabilistic gradient-based meta-learning algorithm that models the uncertainty arising from the few-shot setting with a low-dimensional latent variable. We experimentally evaluate our framework on a simulated reaching and a real-robot picking task using 400 simulated robots generated by varying the physical parameters of an existing set of robotic platforms. Our results show that the proposed method can successfully adapt a trained policy to different robotic platforms with novel physical parameters and the superiority of our meta-learning algorithm compared to state-of-the-art methods for the introduced few-shot policy adaptation problem.

Bat flight has been an underdeveloped area of bio-inspired robotics because of the vast complexities of biological bat flight and the over 40 degrees of freedom present in their bodies. The robotic flapping system Bat Bot (B2) has been shown to exhibit fundamental properties of biological bat flight with its articulated wings, its deformable membrane, and its controllable hindlimbs. However, the system is limited in performance by its relatively large mass for the thrust it produces. In an effort to further pursue this important area of flapping flight, we have made several important hardware improvements to the system based on biological inspiration. These include passive wrist joints to reduce negative lift in the upstroke and a novel elastic fiber membrane to mimic the anistropic nature of bat skin for performance and durability. The redesigned flapping mechanism and structure have reduced the weight by 22%, increased the flapping amplitude, lowered mechanical slackness, and improved mass distribution. These hardware improvements are functional together in free-flight tests. This new system Bat Bot 2.0 (B2.0) provides insights into the important elements of design of bat robots, and it brings the goal of complex bat flight maneuvers closer to reality.

This paper introduces a new class of soft reconfigurable robot: balloon animal robots. The balloon animal robot consists of a closed volume inflatable tube which can be reconfigured into structures of varying topology by a collective of simple sub-unit robots. The robotic sub-units can (1) drive along the length of the tube to localize a joint, (2) create pinch points that locally reduce the bending stiffness of the tube to form a joint, and (3) selectively mechanically couple to one another through cable driven actuators to create nodes of the structure. In this work we introduce the hardware necessary to construct the robot, present experiments to guide the hardware design, and formulate an algorithm using graph theory to calculate the number of nodes and node connections needed to form different 2D shapes. Finally, we demonstrate the system with two active nodes and four passive nodes forming multiple 2D shapes from the same hardware.

Robust visual localization in traffic scenes is a fundamental problem for self-driving vehicles. However, it is still challenging to achieve accurate localization performance because of drastic viewpoint and illumination changes. To address the issues, we design a novel monocular localization framework based on a light-weight prior map, called BSP-MonoLoc, which leverages the 2D semantic primitives from the monocular images and the 3D basic semantic primitives from the prior map. These primitives are commonly available but lack of distinctive signature. To effectively make associations between the 2D and 3D primitives and refine the vehicle’s pose, we adopt an iterative optimization method, where an efficient hierarchical sample strategy is designed to give a good initial prediction for the associations and the pose. Experimental results on the KAIST dataset and our dataset demonstrate the proposed method can achieve high localization accuracy and run at a real-time performance.

Scene recognition is a fundamental task in robotic perception. For human beings, scene recognition is reasonable because they have abundant object knowledge of the real world. The idea of transferring prior object knowledge from humans to scene recognition is significant but still less exploited. In this paper, we propose to utilize meaningful object representations for indoor scene representation. First, we utilize an improved object model (IOM) as a baseline that enriches the object knowledge by introducing a scene parsing algorithm pretrained on the ADE20K dataset with rich object categories related to the indoor scene. To analyze the object co-occurrences and pairwise object relations, we formulate the IOM from a Bayesian perspective as the Bayesian object relation model (BORM). Meanwhile, we incorporate the proposed BORM with the PlacesCNN model as the combined Bayesian object relation model (CBORM) for scene recognition and significantly outperforms the state-of-the-art methods on the reduced Places365 dataset, and SUN RGB-D dataset without retraining, showing the excellent generalization ability of the proposed method. Code can be found at https://github.com/FreeformRobotics/BORM.

LiDAR-only object detection is essential for autonomous driving systems and is a challenging problem. For the representation of a bird’s eye view LiDAR point-cloud, this paper proposes a single-stage object detector. The detector can output classification information and accurate positioning information for multi-category objects. In this paper, the detector’s design methods are detailed from a bird’s eye view LiDAR point-cloud encoding, network design, data augmentation, etc. The detector was evaluated on three challenging datasets: KITTI, nuScenes and Waymo. The experimental results demonstrated that the proposed detector can accurately achieve object detection tasks and the detection speed can reach 26.9 FPS. Both the precision and the speed can meet the requirements of most autonomous driving scenarios.

We propose a 2D path planning algorithm in a non-convex workspace defined as a sequence of connected convex polytopes. The reference path is parameterized as a B-spline curve, which is guaranteed to entirely remain within the workspace by exploiting the local convexity property and by formulating linear constraints on the control points of the B-spline. The novelties of the paper lie in the use of the equivalent Bézier representation of the B-spline curve, which significantly reduces the conservatism in the local convexity bound and in the integration of these constraints into a convex quadratic optimization problem, which minimizes the curve length. The algorithm is successfully validated in both simulations and experiments, by providing obstacle-free reference paths on real occupancy grid maps obtained from the laser scan data of a mobile robot platform.

Optoelectronic tweezer-driven microrobots (OETdMs) are a versatile micromanipulation technology based on the application of light induced dielectrophoresis to move small dielectric structures (microrobots) across a photoconductive substrate. The microrobots in turn can be used to exert forces on secondary objects and carry out a wide range of micromanipulation operations, including collecting, transporting and depositing microscopic cargos. In contrast to alternative (direct) micromanipulation techniques, OETdMs are relatively gentle, making them particularly well suited to interacting with sensitive objects such as biological cells. However, at present such systems are used exclusively under manual control by a human operator. This limits the capacity for simultaneous control of multiple microrobots, reducing both experimental throughput and the possibility of cooperative multi-robot operations. In this article, we describe an approach to automated targeting and path planning to enable open-loop control of multiple microrobots. We demonstrate the performance of the method in practice, using microrobots to simultaneously collect, transport and deposit silica microspheres. Using computational simulations based on real microscopic image data, we investigate the capacity of microrobots to collect target cells from within a dissociated tissue culture. Our results indicate the feasibility of using OETdMs to autonomously carry out micromanipulation tasks within complex, unstructured environments.

During the last decades, the research endeavours on autonomous driving found great resonance in Advanced Driver-Assistance Solutions that equipped the contemporary civilian vehicles and significantly boosted their driver-less mobility. The existing applications are mostly focused on urban scenarios where signs, road lanes and markers are well defined and ordered favouring the motion of the vehicles whilst, less attention has been paid to the semi-structured and rural environments where traffic infrastructure is scarce. The paper at hand introduces a holistic framework for autonomous vehicles navigation in semi-structured environments. Semantic cues fused with geometrical information of LiDAR data are used for road detection and tracking. OpenStreetMaps are employed as a rough route planner, the waypoints of which are rectified via a probability distribution function over the visible area of vehicle’s vicinity. Thus, vehicle’s localization is obtained by Normal Distribution Transform (NDT) SLAM, where the covariance of egomotion estimation is obtained by processing short-term 3D maps, fused with GPS measurements by means of an extended Kalman filter. Local planning and execution of vehicle’s motion is applied on local cost maps formulated by the union of 2D laser readings and the detected road boundaries fitted through Bézier curves. The complete framework has been evaluated with the aid of a real Autonomous Guided Vehicle in a constrained semi-structured urban area, exhibiting robust navigation performance.

Numerous autonomous navigation systems have been proposed, so far, for use in walking environments. Of these, systems that do not rely on high-definition maps and precise localization are cheaper to maintain and easier to implement in unknown outdoor environments. In these systems, road-following navigation using road boundaries is commonly used. In outdoor environments, however, the road boundary is indistinct in some cases. Therefore, it is necessary to infer an area recommended by traffic rules and navigate based on this inference. This paper proposes an autonomous navigation system that is independent of the road boundaries by using a driving recommendation map. In this system, the driving recommendation degree is inferred by deep learning-based semantic segmentation using automatically labeled training data, and a map that probabilistically represents the driving recommendation degree is generated using this information. The proposed autonomous navigation system has demonstrated its usefulness in an environment with indistinct road boundaries. The demonstration shows the proposed autonomous navigation system can plan a path based on driving recommendation degree and properly drive in a scene where the road boundary is indistinct.

In this paper, we introduce a complete system for autonomous flight of quadrotors in dynamic environments with onboard sensing. Extended from existing work, we develop an occlusion-aware dynamic perception method based on depth images, which classifies obstacles as dynamic and static. For representing generic dynamic environment, we model dynamic objects with moving ellipsoids and fuse static ones into an occupancy grid map. To achieve dynamic avoidance, we design a planning method composed of modified kinodynamic path searching and gradient-based optimization. The method leverages manually constructed gradients without maintaining a signed distance field (SDF), making the planning procedure finished in milliseconds. We integrate the above methods into a customized quadrotor system and thoroughly test it in real-world experiments, verifying its effective collision avoidance in dynamic environments.

In many robotic tasks, such as autonomous drone racing, the goal is to travel through a set of waypoints as fast as possible. A key challenge for this task is planning the timeoptimal trajectory, which is typically solved by assuming perfect knowledge of the waypoints to pass in advance. The resulting solution is either highly specialized for a single-track layout, or suboptimal due to simplifying assumptions about the platform dynamics. In this work, a new approach to near-time-optimal trajectory generation for quadrotors is presented. Leveraging deep reinforcement learning and relative gate observations, our approach can compute near-time-optimal trajectories and adapt the trajectory to environment changes. Our method exhibits computational advantages over approaches based on trajectory optimization for non-trivial track configurations. The proposed approach is evaluated on a set of race tracks in simulation and the real world, achieving speeds of up to 60kmh−1 with a physical quadrotor.