ICRA 2022 Papers

3D meshes offer a computationally efficient but still quite accurate path to the understanding of a robot's environment. While mesh reconstructions are often employed in indoor regions where regular planar surfaces dominate the scene, their use in urban outdoor environments has been under-explored. This is as outdoor urban environments produce a significant contrast between preserving discontinuities between different objects and maintaining smoothness in the solution. When coupled with the natural sparse nature of meshes this presents a significant challenge to their optimisation. Motivated by these challenges and leveraging insights from computer graphics, we firstly balance previously introduced real-time definitions of variational smoothing on meshes with their ‘canonical’ definitions. In doing so we introduce a novel Delaunay-Voronoi formulation for real-time mesh smoothing that allows for the accumulation of information away from the triangular edges. This allows for the use of more powerful non-convex regularisers that are able to more finely balance smoothness and object discontinuities and showcase more faithful reconstructions of the urban outdoor scene. In doing so the benefits of non-convex regularisers promised in the ‘every-pixel’ scenarios can now be inherited by sparse mesh formulations.

Unmanned underwater vehicles are crucial for deep-sea exploration and inspection without imposing any danger to human life due to extreme environmental conditions. But, designing a robust controller that can cope with model uncertainties, external disturbances, and time delays for such vehicles is a challenge. This paper implements a sliding mode position control algorithm with a time-delay estimation term to a remotely operated underwater vehicle to deal with disturbances, such as waves, and time delays. The controller is implemented on an underwater vehicle (BlueRov) and compared with a proportional-integral-derivative (PID) controller in a wave tank with different disturbances and when there exist delays within the communication channel. The experimental results show that the proposed control method provides better performance than the conventional PID in the presence of extreme disturbances with less control efforts.

Learning from demonstrations in the wild (e.g. YouTube videos) is a tantalizing goal in imitation learning. However, for this goal to be achieved, imitation learning algorithms must deal with the fact that the demonstrators and learners may have bodies that differ from one another. This condition — “embodiment mismatch” — is ignored by many recent imitation learning algorithms. Our proposed imitation learning technique, SILEM (Skeletal feature compensation for Imitation Learning with Embodiment Mismatch), addresses a particular type of embodiment mismatch by introducing a learned affine transform to compensate for differences in the skeletal features obtained from the learner and expert. We create toy domains based on PyBullet's HalfCheetah and Ant to assess SILEM's benefits for this type of embodiment mismatch. We also provide qualitative and quantitative results on more realistic problems — teaching simulated humanoid agents, including Atlas from Boston Dynamics, to walk by observing human demonstrations.

Prior work on 6-DoF object pose estimation has largely focused on instance-level processing, in which a textured CAD model is available for each object being detected. Category-level 6- DoF pose estimation represents an important step toward developing robotic vision systems that operate in unstructured, real-world scenarios. In this work, we propose a single-stage, keypoint-based approach for category-level object pose estimation that operates on unknown object instances within a known category using a single RGB image as input. The proposed network performs 2D object detection, detects 2D keypoints, estimates 6- DoF pose, and regresses relative bounding cuboid dimensions. These quantities are estimated in a sequential fashion, leveraging the recent idea of convGRU for propagating information from easier tasks to those that are more difficult. We favor simplicity in our design choices: generic cuboid vertex coordinates, single-stage network, and monocular RGB input. We conduct extensive experiments on the challenging Objectron benchmark, outperforming state-of-the-art methods on the 3D IoU metric (27.6% higher than the MobilePose single-stage approach and 7.1 % higher than the related two-stage approach).

This paper proposes a novel motion estimation algorithm using WiFi networks and IMU sensor data in large uncontrolled environments, dubbed “WiFi Structure-from-Motion” (WiFi SfM). Given smartphone sensor data through day-to-day activities from a single user over a month, our WiFi SfM algorithm estimates smartphone motion tra-jectories and the structure of the environment represented as a WiFi radio map. The approach 1) establishes frame-to-frame correspondences based on WiFi fingerprints while exploiting our repetitive behavior patterns; 2) aligns trajectories via bundle adjustment; and 3) trains a self-supervised neural network to extract further motion constraints. We have col-lected 235 hours of smartphone data, spanning 38 days of daily activities in a university campus. Our experiments demonstrate the effectiveness of our approach over the competing methods with qualitative evaluations of the estimated motions and quantitative evaluations of indoor localization accuracy based on the reconstructed WiFi radio map. The WiFi SfM technology will potentially allow digital mapping companies to build better radio maps automatically by asking users to share WiFi/IMU sensor data in their daily activities.

We propose factor graph optimization for simultaneous planning, control, and trajectory estimation for collision-free navigation of autonomous systems in environments with moving objects. The proposed online probabilistic motion planning and trajectory estimation navigation technique generates optimal collision-free state and control trajectories for autonomous vehicles when the obstacle motion model is both unknown and known. We evaluate the utility of the algorithm to support future autonomous robotic space missions.

Soft robots can be incredibly robust and safe but typically fail to match the strength and precision of rigid robots. This dichotomy between soft and rigid is recently starting to break down, with emerging research interest in hybrid soft-rigid robots. In this work, we draw inspiration from Nature, which achieves the best of both worlds by coupling soft and rigid tissues-like muscle and bone-to produce biological systems capable of both robustness and strength. We present foundational, general-purpose pipelines to simulate and fabricate cable-driven soft-rigid robots with embedded skeletons. We show that robots built using these methods can fluidly mimic biological systems while achieving greater force output and external load resistance than purely soft robots. Finally, we show how our simulation and fabrication pipelines can be leveraged to create more complex robots and do model-based control.

Recently, work on reinforcement learning (RL) for bipedal robots has successfully learned controllers for a variety of dynamic gaits with robust sim-to-real demonstrations. In order to maintain balance, the learned controllers have full freedom of where to place the feet, resulting in highly robust gaits. In the real world however, the environment will often impose constraints on the feasible footstep locations, typically identified by perception systems. Unfortunately, most demonstrated RL controllers on bipedal robots do not allow for specifying and responding to such constraints. This missing control interface greatly limits the real-world application of current RL controllers. In this paper, we aim to maintain the robust and dynamic nature of learned gaits while also respecting footstep constraints imposed externally. We develop an RL formulation for training dynamic gait controllers that can respond to specified touchdown locations. We then successfully demonstrate simulation and sim-to-real performance on the bipedal robot Cassie. In addition, we use supervised learning to induce a transition model for accurately predicting the next touchdown locations that the controller can achieve given the robot's proprioceptive observations. This model paves the way for integrating the learned controller into a full-order robot locomotion planner that robustly satisfies both balance and environmental constraints.

Recent work on sim-to-real learning for bipedal locomotion has demonstrated new levels of robustness and agility over a variety of terrains. However, that work, and most prior bipedal locomotion work, have not considered locomotion under a variety of external loads that can significantly influence the overall system dynamics. In many applications, robots will need to maintain robust locomotion under a wide range of potential dynamic loads, such as pulling a cart or carrying a large container of sloshing liquid, ideally without requiring additional load-sensing capabilities. In this work, we explore the capabilities of reinforcement learning (RL) and sim-to-real transfer for bipedal locomotion under dynamic loads using only proprioceptive feedback. We show that prior RL policies trained for unloaded locomotion fail for some loads and that simply training in the context of loads is enough to result in successful and improved policies. We also compare training specialized policies for each load versus a single policy for all considered loads and analyze how the resulting gaits change to accommodate different loads. Finally, we demonstrate sim-to-real transfer, which is successful but shows a wider sim-to-real gap than prior unloaded work, which points to interesting future research.

Siamese network based trackers have achieved significant progress in visual object tracking. For the sake of speed, they mainly rely on offline training to learn a mono-level feature correlation between a target template and a search region. During the tracking period, they use a fixed strategy to infer target positions over sequences regardless of target states. However, such approaches are vulnerable in case of long-term challenges e.g. large variance, presence of distractors, fast motion, or target disappearing and the like. In this paper, we propose a new tracking framework, referred to as SiamX, by exploiting cross-level Siamese features to learn robust correlations between the target template and search regions, and also adaptive inference strategies to prevent tracking loss and realize fast target re-localization. Extensive experiments on four benchmarks including VOT-2019, LaSOT, GOT-10k, and TrackingNet show our method significantly enhances the tracker's ability to resist variance and interference, and achieve state-of-the-art results at around 50 FPS.

Knowledge of 3-D object shape is of great importance to robot manipulation tasks, but may not be readily available in unstructured environments. While vision is often occluded during robot-object interaction, high-resolution tactile sensors can give a dense local perspective of the object. However, tactile sensors have limited sensing area and the shape representation must faithfully approximate non-contact areas. In addition, a key challenge is efficiently incorporating these dense tactile measurements into a 3-D mapping framework. In this work, we propose an incremental shape mapping method using a GelSight tactile sensor and a depth camera. Local shape is recovered from tactile images via a learned model trained in simulation. Through efficient inference on a spatial factor graph informed by a Gaussian process, we build an implicit surface representation of the object. We demonstrate visuo-tactile mapping in both simulated and real-world experiments, to incrementally build 3-D reconstructions of household objects.

The shape control of deformable linear objects (DLOs) is challenging, since it is difficult to obtain the deformation models. Previous studies often approximate the models in purely offline or online ways. In this paper, we propose a scheme for the shape control of DLOs, where the unknown model is estimated with both offline and online learning. The model is formulated in a local linear format, and approximated by a neural network (NN). First, the NN is trained offline to provide a good initial estimation of the model, which can directly migrate to the online phase. Then, an adaptive controller is proposed to achieve the shape control tasks, in which the NN is further updated online to compensate for any errors in the offline model caused by insufficient training or changes of DLO properties. The simulation and real-world experiments show that the proposed method can precisely and efficiently accomplish the DLO shape control tasks, and adapt well to new and untrained DLOs.

Shape and pose estimation is a critical perception problem for a self-driving car to fully understand its surrounding environment. One fundamental challenge in solving this problem is the incomplete sensor signal (e.g., LiDAR scans), especially for faraway or occluded objects. In this paper, we propose a novel algorithm to address this challenge, which explicitly leverages the sensor signal captured over consecutive time: the consecutive signals can provide more information about an object, including different viewpoints and its motion. By encoding the consecutive signals via a recurrent neural network, not only our algorithm improves the shape and pose estimates, but also produces a labeling tool that can benefit other tasks in autonomous driving research. Specifically, building upon our algorithm, we propose a novel pipeline to automatically annotate high-quality labels for amodal segmentation on images, which are hard and laborious to annotate manually. Our code and data will be made publicly available.

Despite advances in a diversity of environments, snake robots are still far behind snakes in traversing complex 3-D terrain with large obstacles. This is due to a lack of understanding of how to control 3-D body bending to push against terrain features to generate and control propulsion. Biological studies suggested that generalist snakes use contact force sensing to adjust body bending in real time to do so. However, studying this sensory-modulated force control in snakes is challenging, due to a lack of basic knowledge of how their force sensing organs work. Here, we take a robophysics approach to make progress, starting by developing a snake robot capable of 3-D body bending with contact force sensing to enable systematic locomotion experiments and force measurements. Through two development and testing iterations, we created a 12-segment robot with 36 piezo-resistive sheet sensors distributed on all segments with compliant shells with a sampling frequency of 30 Hz. The robot measured contact forces while traversing a large obstacle using vertical bending with high repeatability, achieving the goal of providing a platform for systematic experiments. Finally, we explored model-based calibration considering the viscoelastic behavior of the piezo-resistive sensor, which will for useful for future studies.

Perception and portability are critical issues for wearable gloves in hand assistive engineering. However, available wearable gloves either lack flexible sensing or are bulky. In this paper, we present a tendon-driven lightweight wearable glove with soft joint sensing, Sen-Glove. Sen-Glove is equipped with 14 soft strain sensors, which enables full bending motion monitoring of 14 joints of five fingers and greatly reduces the weight of the glove. Besides, modular design makes Sen-Glove more compact and weighs 161g in total, reducing the burden on hand. A series of mechanical tests are conducted to evaluate the characteristics of Sen-Glove. Experimental results show that Sen-Glove can withstand 500 bending cycles, assist the subject in grasping 21 multi-scale objects, and recognize 11 gestures. The classification accuracy of 11 different gestures reaches 98.6 %, which verifies the efficacy of the strain sensors.

We propose a semi-supervised learning framework for monocular depth estimation. Compared to existing semi-supervised learning methods, which inherit limitations of both sparse supervised and unsupervised loss functions, we achieve the complementary advantages of both loss functions, by building two separate network branches for each loss and distilling each other through the mutual distillation loss function. We also present to apply different data augmentation to each branch, which improves the robustness. We conduct experiments to demonstrate the effectiveness of our framework over the latest methods and provide extensive ablation studies.

In this paper, we present a semi-autonomous teleoperation framework for a pick-and-place task using an RGB-D sensor. In particular, we assume that the target object is located in a cluttered environment where both prehensile grasping and non-prehensile manipulation are combined for efficient teleoperation. A trajectory-based reinforcement learning is utilized for learning the non-prehensile manipulation to rearrange the objects for enabling direct grasping. From the depth image of the cluttered environment and the location of the goal object, the learned policy can provide multiple options of non-prehensile manipulation to the human operator. We carefully design a reward function for the rearranging task where the policy is trained in a simulational environment. Then, the trained policy is transferred to a real-world and evaluated in a number of real-world experiments with the varying number of objects where we show that the proposed method outperforms manual keyboard control in terms of the time duration for the grasping.

Object rearrangement has recently emerged as a key competency in robot manipulation, with practical solutions generally involving object detection, recognition, grasping and high-level planning. Goal-images describing a desired scene configuration are a promising and increasingly used mode of instruction. A key outstanding challenge is the accurate inference of matches between objects in front of a robot, and those seen in a provided goal image, where recent works have struggled in the absence of object-specific training data. In this work, we explore the deterioration of existing methods' ability to infer matches between objects as the visual shift between observed and goal scenes increases. We find that a fundamental limitation of the current setting is that source and target images must contain the same instance of every object, which restricts practical deployment. We present a novel approach to object matching that uses a large pre-trained vision-language model to match objects in a cross-instance setting by leveraging semantics together with visual features as a more robust, and much more general, measure of similarity. We demonstrate that this provides considerably improved matching performance in cross-instance settings, and can be used to guide multi-object rearrangement with a robot manipulator from an image that shares no object instances with the robot's scene. Our code is available at https://github.com/applied-ai-lab/object_matching.

Underwater image enhancement has become an attractive topic as a significant technology in marine engi-neering and aquatic robotics. However, the limited number of datasets and imperfect hand-crafted ground truth weaken its robustness to unseen scenarios, and hamper the application to high-level vision tasks. To address the above limitations, we develop an efficient and compact enhancement network in collaboration with a high-level semantic-aware pretrained model, aiming to exploit its hierarchical feature representation as an auxiliary for the low-level underwater image enhance-ment. Specifically, we tend to characterize the shallow layer features as textures while the deep layer features as structures in the semantic-aware model, and propose a multi-path Contextual Feature Refinement Module (CFRM) to refine features in multiple scales and model the correlation between different features. In addition, a feature dominative network is devised to perform channel-wise modulation on the aggregated texture and structure features for the adaptation to different feature patterns of the enhancement network. Extensive experiments on benchmarks demonstrate that the proposed algorithm achieves more appealing results and outperforms state-of-the-art meth-ods by large margins. We also apply the proposed algorithm to the underwater salient object detection task to reveal the favorable semantic-aware ability for high-level vision tasks.