IROS 2021 Papers

Adopting the ideal reliable machine model, the throughput of a lithography machine can be given as the reciprocal of the operation time. This time can be defined at the die level where the actual exposure process takes place as the time unit per die. A closer look at the motion profiles, namely step-and-scan trajectories, suggests that a multi-disciplinary design optimization should be involved when such profiles are selected or designed. Being the reference motion used, the step-and-scan trajectories not only affect the machine performance, but also affect its throughput and to an extent the die yield as well. Structural vibration, and thermal loading at the actuators due to friction and repetitive motion may build up because of following the reference motion. Moreover, since the exposure process and equipment are synchronized with the reference motion, deformation and thermal stress may affect the reticle, the wafer and the projection elements if the exposure high-energy duration and frequency are not taken into consideration while designing the reference motion. From dynamics point of view, reference motion with higher-order derivatives enhances the tracking performance of the machine, however, its operational cost is usually overlooked. In this paper, we present a case-study that outlines the aforementioned aspects using three step-and-scan profiles of the same order. We conclude by posing the following research question: what is the best combination of orders of the step and the scan trajectories that jointly meet the desired performance and operating conditions?

Different health-monitoring techniques were considered in the literature to enhance the safety and stability of Connected Autonomous Vehicle (CAV) platoons. The health-monitoring processes include fault detection, localization, and mitigation. It is evident that mitigating these faults is faster and more reliable if the fault structure is known. To this end, we consider classifying the fault class using supervised machine learning. We first model a heterogeneous CAV platoon with three different common faults separately. These faults are bounded actuator disturbances (namely, engine bearing knock), False Data Injection (FDI) attack, and communication time delay. We consider two supervised machine learning classifiers, the first classifier determines whether the fault is bounded disturbances or communication delay, and the second classifier determines whether the disturbances are in the physical or cyber layer. We have compared four machine learning techniques for each classifier, Support Vector Machine (SVM), Naive Bayes (NB), Quadratic Discriminant (QD), and K-Nearest Neighbors (KNN). The classifiers are trained firstly on the simulation model, then are tested on a different set of observations and tested experimentally on a platoon of three autonomous robots. The highest accuracy was achieved by considering SVM for the first classifier and QD for the second classifier. The overall classification accuracy achieved is 96.8% for the simulation test and 92.1% for the experiment.

The ability to simultaneously distinguish objects, materials, and their associated physical properties is one fundamental function of the sense of touch. Recent advances in the development of tactile sensors and machine learning techniques allow more accurate and complex modelling of robotic tactile sensations. However, many state-of-the-art (SotA) approaches focus solely on constructing black-box models to achieve ever higher classification accuracy and fail to adapt across sensors with unique spatial-temporal data formats. In this work, we propose an Explainable and Sensor-Adaptable Recurrent Networks (ExSARN) model for tactile texture representation. The ExSARN model consists of a two-stage recurrent networks fed by a sensor-specific header network. The first stage recurrent network emulates our human touch receptors and decouples sensor-specific tactile sensations into different frequency response bands, while the second stage codes the overall temporal signature as a variational recurrent autoen-coder. We infuse the latent representation with ternary labels to qualitatively represent texture properties (e.g. roughness and stiffness), which facilitates representation learning and provide explainability to the latent space. The ExSARN model is tested on texture datasets collected with two different tactile sensors. Our results show that the proposed model not only achieves higher accuracy, but also provides adaptability across sensors with different sampling frequencies and data formats. The addition of the crudely obtained qualitative property labels offers a practical approach to enhance the interpretability of the latent space, facilitate property inference on unseen materials, and improve the overall performance of the model.

In many situations it is either impossible or impractical to develop and evaluate agents entirely on the target domain on which they will be deployed. This is particularly true in robotics, where doing experiments on hardware is much more arduous than in simulation. This has become arguably more so in the case of learning-based agents. To this end, considerable recent effort has been devoted to developing increasingly realistic and higher fidelity simulators. However, we lack any principled way to evaluate how good a "proxy domain" is, specifically in terms of how useful it is in helping us achieve our end objective of building an agent that performs well in the target domain. In this work, we investigate methods to address this need. We begin by clearly separating two uses of proxy domains that are often conflated: 1) their ability to be a faithful predictor of agent performance and 2) their ability to be a useful tool for learning. In this paper, we attempt to clarify the role of proxy domains and establish new proxy usefulness (PU) metrics to compare the usefulness of different proxy domains. We propose the relative predictive PU to assess the predictive ability of a proxy domain and the learning PU to quantify the usefulness of a proxy as a tool to generate learning data. Furthermore, we argue that the value of a proxy is conditioned on the task that it is being used to help solve. We demonstrate how these new metrics can be used to optimize parameters of the proxy domain for which obtaining ground truth via system identification is not trivial.

This paper presents an offset-free model predictive controller for fast and accurate control of a spherical soft robotic arm. In this control scheme, a linear model is combined with an online disturbance estimation technique to systematically compensate model deviations. Dynamic effects such as material relaxation resulting from the use of soft materials can be addressed to achieve offset-free tracking. The tracking error can be reduced by 35% when compared to a standard model predictive controller without a disturbance compensation scheme. The improved tracking performance enables the realization of a ball catching application, where the spherical soft robotic arm can catch a ball thrown by a human.

In the era of self-driving vehicles, state estimation has 3 main contradictory requirements, such as accuracy, robustness, and cost-effectiveness. To satisfy all of them, the integration of the wheel encoder measurements is a proper choice besides the generally applied GNSS, inertial and visual-odometry methods. The wheel odometry is a robust and cost-effective method, but the accuracy of the estimation is limited by the proper knowledge of the vehicle parameter values. However, the calibration of the nonlinear odometry model in the presence of noise remains an open problem in the context of autonomous vehicles yet. This paper presents an algorithm that takes advantage of the assumption that more measurements are available in a self-driving vehicle for accurate parameter estimation. With the proposed architecture, the measurements with distortion effects are detected, and also the noise is corrected to reach unbiased model calibration. The performance of the developed algorithm and the accuracy of the parameter estimation are demonstrated with detailed validation and test with a real vehicle.

We address the manipulation task of retrieving a target object from a cluttered shelf. When the target object is hidden, the robot must search through the clutter for retrieving it. Solving this task requires reasoning over the likely locations of the target object. It also requires physics reasoning over multi-object interactions and future occlusions. In this work, we present a data-driven hybrid planner for generating occlusion-aware actions in closed-loop. The hybrid planner explores likely locations of the occluded target object as predicted by a learned distribution from the observation stream. The search is guided by a heuristic trained with reinforcement learning to act on observations with occlusions. We evaluate our approach in different simulation and real-world settings (video available on https://youtu.be/dY7YQ3LUVQg). The results validate that our approach can search and retrieve a target object in near real time in the real world while only being trained in simulation.

Micro Air Vehicles (MAVs) are increasingly being used for complex or hazardous tasks in enclosed and cluttered environments such as surveillance or search and rescue. With this comes the necessity for sensors that can operate in poor visibility conditions to facilitate with navigation and avoidance of objects or people. Radar sensors in particular can provide more robust sensing of the environment when traditional sensors such as cameras fail in the presence of dust, fog or smoke. While extensively used in autonomous driving, miniature FMCW radars on MAVs have been relatively unexplored. This study aims to investigate to what extent this sensor is of use in these environments by employing traditional signal processing such as multi-target tracking and velocity obstacles. The viability of the solution is evaluated with an implementation on board a MAV by running trial tests in an indoor environment containing obstacles and by comparison with a human pilot, demonstrating the potential for the sensor to provide a more robust sense and avoid function in fully autonomous MAVs.

Accurate perception of the surrounding scene is helpful for robots to make reasonable judgments and behaviours. Therefore, developing effective scene representation and recognition methods are of significant importance in robotics. Currently, a large body of research focuses on developing novel auxiliary features and networks to improve indoor scene recognition ability. However, few of them focus on directly constructing object features and relations for indoor scene recognition. In this paper, we analyze the weaknesses of current methods and propose an Object-to-Scene (OTS) method, which extracts object features and learns object relations to recognize indoor scenes. The proposed OTS first extracts object features based on the segmentation network and the proposed object feature aggregation module (OFAM). Afterwards, the object relations are calculated and the scene representation is constructed based on the proposed object attention module (OAM) and global relation aggregation module (GRAM). The final results in this work show that OTS successfully extracts object features and learns object relations from the segmentation network. Moreover, OTS outperforms the state-of-the-art methods by more than 2% on indoor scene recognition without using any additional streams. Code is publicly available at: https://github.com/FreeformRobotics/OTS.

We propose a novel object-augmented RGB-D SLAM system that is capable of constructing a consistent object map and performing relocalisation based on centroids of objects in the map. The approach aims to overcome the view dependence of appearance-based relocalisation methods using point features or images. During the map construction, we use a pre-trained neural network to detect objects and estimate 6D poses from RGB-D data. An incremental probabilistic model is used to aggregate estimates over time to create the object map. Then in relocalisation, we use the same network to extract objects-of-interest in the ‘lost’ frames. Pairwise geometric matching finds correspondences between map and frame objects, and probabilistic absolute orientation followed by application of iterative closest point to dense depth maps and object centroids gives relocalisation. Results of experiments in desktop environments demonstrate very high success rates even for frames with widely different viewpoints from those used to construct the map, significantly outperforming two appearance- based methods.

Object picking with two-fingered grippers is widely used in practice. However, the deformability and slipperiness of the target object still remain a challenge, and not resolving them might lead to breaking or dropping of the grasped objects. To prevent such instances, tactile sensing plays an important role because it can directly detect even the subtle changes that occur during grasping. Mechanoreceptors in the human skin detect such events by the change in the skin shape and/or vibration. Using a similar approach, a combined deformation and slip detection system using a distributed 3axis tactile information with various time-scales is proposed. Specifically, the tactile information includes the z-axis data, which denotes the deformation of the skin perpendicular to the finger’s surface and the x- and y-axes, which measure deformations tangential to the surface. The perpendicular and tangential tactile information are used to determine the deformation and slip, respectively. The system is based on a multilayer perceptron (MLP) that outputs detection results from a 3-axis tactile information. Results showed that, the perpendicular and tangential tactile information with an appropriate timescale were effective for deformation and slip detection with over 89% and 95% recognition rates, respectively, measured for 40 different objects. Moreover, 195 out of 200 real-time untrained grasping states were successful detected. Finally, 10 untrained objects were successfully picked.

Object models are highly useful for robots as they enable tasks such as detection, pose estimation and manipulation. However, models are not always easily available, especially in real-world domains of operation such as peoples’ homes. This work presents a pipeline to generate high-quality object reconstructions from human in-hand manipulation to alleviate the necessity of specialised or expensive hardware. Missing data, due to occlusion or unseen sides, is explicitly handled by incorporating shape completion. We demonstrate the usability of the reconstructions by applying a model-based as well as a CNN-based object pose estimator that is trained on synthetic images by employing state-of-the-art texture synthesis. Using our pipeline to cheaply generate object models and synthetic RGB images for training, we achieve competitive performance compared to baselines that require an elaborate set-up to construct models or large amounts of annotated data. Object grasping is also enabled by learning with the reconstructions in simulation, then executing with a real robot. These evaluations show that our reconstructions are comparable to those made under near-perfect conditions and enable 6D object pose estimation as well as real-world grasping.

Stereo reconstruction models trained on small images do not generalize well to high-resolution data. Training a model on high-resolution image size faces difficulties of data availability and is often infeasible due to limited computing resources. In this work, we present the Occlusion-aware Recurrent binocular Stereo matching (ORStereo), which deals with these issues by only training on available low disparity range stereo images. ORStereo generalizes to unseen high-resolution images with large disparity ranges by formulating the task as residual updates and refinements of an initial prediction. ORStereo is trained on images with disparity ranges limited to 256 pixels, yet it can operate 4K-resolution input with over 1000 disparities using limited GPU memory. We test the model’s capability on both synthetic and real-world high-resolution images. Experimental results demonstrate that ORStereo achieves comparable performance on 4K-resolution images compared to state-of-the-art methods trained on large disparity ranges. Compared to the baseline methods that are only trained on low-resolution images, our method has 60% or less error on 4K-resolution images.

The successful performance of any system is dependant on the hardware of the agent, which is typically immutable during RL training. In this work, we present ORCHID (Optimisation of Robotic Control and Hardware In Design) which allows for truly simultaneous optimisation of hardware and control parameters in an RL pipeline. We show that by forming a complex differential path through a trajectory rollout we can leverage a vast amount of information from the system that was previously lost in the ‘black-box’ environment. Combining this with a novel hardware-conditioned critic network minimises variance during training and ensures stable updates are made. This allows for refinements to be made to both the morphology and control parameters simultaneously. The result is an efficient and versatile approach to holistic robot design, that brings the final system nearer to true optimality. We show improvements in performance across 4 different test environments with two different control algorithms - in all experiments the maximum performance achieved with ORCHID is shown to be unattainable using only policy updates with the default design. We also show how re-designing a robot using ORCHID in simulation, transfers to a vast improvement in the performance of a real-world robot.

The data loss caused by unreliable network seriously impacts the results of remote visual SLAM systems. From our experiment, a loss of less than 1 second of data can cause a visual SLAM algorithm to lose tracking. We present a novel buffering method, ORBBuf, to reduce the impact of data loss on remote visual SLAM systems. We model the buffering problem as an optimization problem by introducing a similarity metric between frames. To solve the buffering problem, we present an efficient greedy algorithm to discard the frames that have the least impact on the quality of SLAM results. We implement our ORBBuf method on ROS, a widely used middleware framework. Through an extensive evaluation on real-world scenarios and tens of gigabytes of datasets, we demonstrate that our ORBBuf method can be applied to different state-estimation algorithms (DSO and VINS-Fusion), different sensor data (both monocular images and stereo images), different scenes (both indoor and outdoor), and different network environments (both WiFi networks and 4G networks). Our experimental results indicate that the network losses indeed affect the SLAM results, and our ORBBuf method can reduce the RMSE up to 50 times comparing with the Drop-Oldest and Random buffering methods.

Humans have mental models that allow them to plan, experiment, and reason in the physical world. How should an intelligent agent go about learning such models? In this paper, we will study if models of the world learned in an open-ended physics environment, without any specific tasks, can be reused for downstream physics reasoning tasks. To this end, we build a benchmark Open-ended Physics Environment (OPEn) and also design several tasks to test learning representations in this environment explicitly. This setting reflects the conditions in which real agents (i.e. rolling robots) find themselves, where they may be placed in a new kind of environment and must adapt without any teacher to tell them how this environment works. This setting is challenging because it requires solving an exploration problem in addition to a model building and representation learning problem. We test several existing RL-based exploration methods on this benchmark and find that an agent using unsupervised contrastive learning for representation learning, and impact-driven learning for exploration, achieved the best results. However, all models still fall short in sample efficiency when transferring to the downstream tasks. We expect that OPEn will encourage the development of novel rolling robot agents that can build reusable mental models of the world that facilitate many tasks.

The control of a robot for manipulation tasks generally relies on object detection and pose estimation. An attractive alternative is to learn control policies directly from raw input data. However, this approach is time-consuming and expensive since learning the policy requires many trials with robot actions in the physical environment. To reduce the training cost, the policy can be learned in simulation with a large set of synthetic images. The limit of this approach is the domain gap between the simulation and the robot workspace. In this paper, we propose to learn a policy for robot reaching movements from a single image captured directly in the robot workspace from a camera placed on the end-effector (a hand-eye camera). The idea behind the proposed policy learner is that view changes seen from the hand-eye camera produced by actions in the robot workspace are analogous to locating a region-of-interest in a single image by performing sequential object localisation. This similar view change enables training of object reaching policies using reinforcement-learning-based sequential object localisation. To facilitate the adaptation of the policy to view changes in the robot workspace, we further present a dynamic filter that learns to bias an input state to remove irrelevant information for an action decision. The proposed policy learner can be used as a powerful representation for robotic tasks, and we validate it on static and moving object reaching tasks.

Object detection plays a deep role in visual systems by identifying instances for downstream algorithms. In industrial scenarios, however, a slight change in manufacturing systems would lead to costly data re-collection and human annotation processes to re-train models. Existing solutions such as semi-supervised and few-shot methods either rely on numerous human annotations or suffer low performance. In this work, we explore a novel object detector based on interactive perception (ODIP), which can be adapted to novel domains in an automated manner. By interacting with a grasping system, ODIP accumulates visual observations of novel objects, learning to identify previously unseen instances without humanannotated data. Extensive experiments show ODIP outperforms both the generic object detector and state-of-the-art few-shot object detector fine-tuned in traditional manners. A demo video is provided to further illustrate the idea [1].

Recent advances in object segmentation have demonstrated that deep neural networks excel at object segmentation for specific classes in color and depth images. However, their performance is dictated by the number of classes and objects used for training, thereby hindering generalization to never seen objects or zero-shot samples. To exacerbate the problem further, object segmentation using image frames rely on recognition and pattern matching cues. Instead, we utilize the ‘active’ nature of a robot and their ability to ‘interact’ with the environment to induce additional geometric constraints for segmenting zero-shot samples.In this paper, we present the first framework to segment unknown objects in a cluttered scene by repeatedly ‘nudging’ at the objects and moving them to obtain additional motion cues at every step using only a monochrome monocular camera. We call our framework NudgeSeg. These motion cues are used to refine the segmentation masks. We successfully test our approach to segment novel objects in various cluttered scenes and provide an extensive study with image and motion segmentation methods. We show an impressive average detection rate of over 86% on zero-shot objects.