ICRA 2020 论文列表

Micromanipulation is challenging due to the specific physical effects governing the microworld. Interactive approaches using only visual feedback are limited to the 2D image of the microscope, and have forcibly lower bandwidth. Recently, haptic feedback teleoperation systems have been developed to try to overcome those difficulties. This paper explores the case of an optical tweezers platform coupled to an haptic device providing transparent force feedback. The impact of haptic feedback regarding user dexterity on tactile exploration tasks is studied using 3 μm microbeads and a test bench with micro sized shapes. The results reveal a consistent improvement in both users' trajectory tracking and their control of the contact forces. This also validates the experimental setup which performed reliably on 140 different trials of the evaluation.

Learning Assistance by Demonstration (LAD) is concerned with using demonstrations of a human agent to teach a robot how to assist another human. The concept has previously been used with smart wheelchairs to provide customised assistance to individuals with driving difficulties. A basic premise of this technique is that the learned assistive policy should be able to generalise to environments different than the ones used for training; but this has not been tested before. In this work we evaluate the assistive power and the generalisation capability of LAD using our custom teleoperation and learning system for smart wheelchairs, while seeking to improve it by experimenting with different combinations of dimensionality reduction techniques and machine learning models. Using Autoencoders to reduce the dimension of laserscan data and a Gaussian Process as the learning model, we achieved a 23% improvement in prediction performance against the combination used by the latest work on the field. Using this model to assist a driver exposed to a simulated disability, we observed a 9.8% reduction in track completion times when compared to driving without assistance.

Robotic Hopping is challenging from the perspective of both modeling the dynamics as well as the mechanical design due to the short period of ground contact in which to actuate on the world. Previous work has demonstrated stable hopping on a moving-mass robot, wherein a single spring was utilized below the body of the robot. This paper finds that the addition of a spring in parallel to the actuator greatly improves the performance of moving mass hopping robots. This is demonstrated through the design of a novel one-dimensional hopping robot. For this robot, a rigorous trajectory optimization method is developed using hybrid systems models with experimentally tuned parameters. Simulation results are used to study the effects of a parallel spring on energetic efficiency, stability, and hopping effort. We find that the double-spring model had 2.5x better energy efficiency than the single-spring model, and was able to hop using 40% less peak force from the actuator. Furthermore, the double-spring model produces stable hopping without the need for stabilizing controllers. These concepts are demonstrated experimentally on a novel hopping robot, wherein hop heights up to 40cm were achieved with comparable efficiency and stability.

Wireless Capsule Endoscopy (WCE) has the advantage of reducing the invasiveness and pain of gastrointestinal examinations. In this work, we propose a system aimed at autonomously accelerating and locating the WCE inside the intestine for clinical applications. A rotating magnet controlled by a robotic arm is placed outside the patient's body to actuate the capsule with an internal magnetic ring, and the magnetic fields of the two sources are measured by an external sensor array. The original Multiple Objects Tracking method is improved by combining Normal Vector Fitting, Bézier Curve Gradient, and Spherical Linear Interpolation to estimate the 6-D pose of the WCE from a 5-D pose sequence. In order to close the actuation-localization loop, a strategy is presented to react to different states of the capsule. The proposed method is validated via experiments on phantoms as well as on animal intestines. The localization of the capsule shows an accuracy of 3.5mm in position and 9.4° in orientation, and the average update frequency of the estimated 6-D pose reaches 25Hz.

We present a simple method to quickly explore C-spaces of robotic manipulators and thus facilitate path planning. The method is based on a novel geometrical structure called generalized bur. It is a star-like tree, rooted at a given point in free C-space, with an arbitrary number of guaranteed collision-free edges computed using distance information from the workspace and simple forward kinematics. Generalized bur captures large portions of free C-space, enabling accelerated exploration. The workspace is assumed to be decomposable into a finite set of (possibly overlapping) convex obstacles. When plugged in a suitable RRT-like planning algorithm, generalized burs enable significant performance improvements, while at the same time enabling exact collision-free paths.

Remote characterisation of the environment during physical robot-environment interaction is an important task commonly accomplished in telerobotics. This paper demonstrates how tactile and proximity sensing can be efficiently used to perform automatic crack detection. A custom-designed integrated tactile and proximity sensor is implemented. It measures the deformation of its body when interacting with the physical environment and distance to the environment's objects with the help of fibre optics. This sensor was used to slide across different surfaces and the data recorded during the experiments was used to detect and classify cracks, bumps and undulations. The proposed method uses machine learning techniques (mean absolute value as feature and random forest as classifier) to detect cracks and determine their width. An average crack detection accuracy of 86.46% and width classification accuracy of 57.30% is achieved. Kruskal-Wallis results (p<; 0.001) indicate statistically significant differences among results obtained when analysing only force data, only proximity data and both force and proximity data. In contrast to previous techniques, which mainly rely on visual modality, the proposed approach based on optical fibres is suitable for operation in extreme environments, such as nuclear facilities in which nuclear radiation may damage the electronic components of video cameras.

Impedance controllers are popularly used in the field of lower limb prostheses and exoskeleton development. Such controllers assume the joint to be a spring-damper system described by a discrete set of equilibria and impedance parameters. These parameters are estimated via a least squares optimization that minimizes the difference between the controller's output torque and human joint torque. Other researchers have used perturbation studies to determine empirical values for ankle impedance. The resulting values vary greatly from the prior least squares estimates. While perturbation studies are more credible, they require immense investment. This paper extended the least squares approach to reproduce the results of perturbation studies. The resulting ankle impedance parameters were successfully tested on a powered transfemoral prosthesis, AMPRO II. Further, the paper investigated the effect of multiple equilibria on the least squares estimation and the performance of the impedance controller. Finally, the paper uses the proposed least squares optimization method to estimate knee impedance.

Autonomous urban driving navigation is still an open problem and has ample room for improvement in unknown complex environments and terrible weather conditions. In this paper, we propose a two-stage framework, called IPP-RL, to handle these problems. IPP means an Imitation learning method fusing visual information with the additional steering angle calculated by Pure-Pursuit (PP) method, and RL means using Reinforcement Learning for further training. In our IPP model, the visual information captured by camera can be compensated by the calculated steering angle, thus it could perform well under bad weather conditions. However, imitation learning performance is limited by the driving data severely. Thus we use a reinforcement learning method-Deep Deterministic Policy Gradient (DDPG)-in the second stage training, which shares the learned weights from pretrained IPP model. In this way, our IPP-RL can lower the dependency of imitation learning on demonstration data and solve the problem of low exploration efficiency caused by randomly initialized weights in reinforcement learning. Moreover, we design a more reasonable reward function and use the n-step return to update the critic-network in DDPG. Our experiments on CARLA driving benchmark demonstrate that our IPP-RL is robust to lousy weather conditions and shows remarkable generalization capability in unknown environments on navigation task.

Most of the research effort on image-based place recognition is designed for urban environments. In bucolic environments such as natural scenes with low texture and little semantic content, the main challenge is to handle the variations in visual appearance across time such as illumination, weather, vegetation state or viewpoints. The nature of the variations is different and this leads to a different approach to describing a bucolic scene. We introduce a global image description computed from its semantic and topological information. It is built from the wavelet transforms of the image's semantic edges. Matching two images is then equivalent to matching their semantic edge transforms. This method reaches state-of-the-art image retrieval performance on two multi-season environment-monitoring datasets: the CMU-Seasons and the Symphony Lake dataset. It also generalizes to urban scenes on which it is on par with the current baselines NetVLAD and DELF.

Although autonomous control of robotic manipulators has been studied for several decades, they are not commonly used in safety-critical applications due to lack of safety and performance guarantees - many of them concerning the modulation of interaction forces. This paper presents a mechanical probing strategy for estimating the environmental impedance parameters of compliant environments, independent a manipulator's controller design, and configuration. The parameter estimates are used in a position-based adaptive force controller to enable control of interaction forces in compliant, stationary, and non-stationary environments. This approach is targeted for applications where the workspace is constrained and non-stationary, and where force control is critical to task success. These applications include surgical tasks involving manipulation of compliant, delicate, moving tissues. Results show fast parameter estimation and successful force modulation that compensates for motion.

This paper proposes a model identification method and evaluation of a force model for multirotor UAVs. The model incorporates propellers' aerodynamics derived from momentum and blade element theories, as well as aerodynamics of the UAV's structure and actuation dynamics. A two-steps identification approach of the model parameters is proposed. The model is identified and evaluated from outdoor experiments with flight speeds exceeding 10m/s.

This paper presents a unique unmanned ground vehicle with a dynamic wheelbase and an adaptive thrust based friction optimization scheme that aids in the traversal of steep slopes and slippery surfaces. The vehicle is capable of adapting itself to the surface topography using an impedance-based stabilization module to minimize the mechanical oscillatory transients induced during its motion. A detailed analysis of its modules has been elucidated in this paper based on the vehicle parameters. The proposed methodologies have been integrated and tested on a customized prototype. Experimental validation and simulation for the proposed modules at various terrain conditions have been carried out to authenticate its performance.

Learning from offline task demonstrations is a problem of great interest in robotics. For simple short-horizon manipulation tasks with modest variation in task instances, offline learning from a small set of demonstrations can produce controllers that successfully solve the task. However, leveraging a fixed batch of data can be problematic for larger datasets and longer-horizon tasks with greater variations. The data can exhibit substantial diversity and consist of suboptimal solution approaches. In this paper, we propose Implicit Reinforcement without Interaction at Scale (IRIS), a novel framework for learning from large-scale demonstration datasets. IRIS factorizes the control problem into a goal-conditioned low-level controller that imitates short demonstration sequences and a high-level goal selection mechanism that sets goals for the low-level and selectively combines parts of suboptimal solutions leading to more successful task completions. We evaluate IRIS across three datasets, including the RoboTurk Cans dataset collected by humans via crowdsourcing, and show that performant policies can be learned from purely offline learning. Additional results at https://sites.google.com/stanford.edu/iris/.

Feature descriptor matching is a critical step is many computer vision applications such as image stitching, image retrieval and visual localization. However, it is often affected by many practical factors which will degrade its performance. Among these factors, illumination variations are the most influential one, and especially no previous descriptor learning works focus on dealing with this problem. In this paper, we propose IF-Net, aimed to generate a robust and generic descriptor under crucial illumination changes conditions. We find out not only the kind of training data important but also the order it is presented. To this end, we investigate several dataset scheduling methods and propose a separation training scheme to improve the matching accuracy. Further, we propose a ROI loss and hard-positive mining strategy along with the training scheme, which can strengthen the ability of generated descriptor dealing with large illumination change conditions. We evaluate our approach on public patch matching benchmark and achieve the best results compared with several state-of-the-arts methods. To show the practicality, we further evaluate IF-Net on the task of visual localization under large illumination changes scenes, and achieves the best localization accuracy.

A robot operating in the world constantly receives information about its environment in the form of new measurements at every time step. Smoothing-based estimation methods seek to optimize for the most likely robot state estimate using all measurements up till the current time step. Existing methods solve for this smoothing objective efficiently by framing the problem as that of incremental unconstrained optimization. However, in many cases observed measurements and knowledge of the environment is better modeled as hard constraints derived from real-world physics or dynamics. A key challenge is that the new optimality conditions introduced by the hard constraints break the matrix structure needed for incremental factorization in these incremental optimization methods. Our key insight is that if we leverage primal-dual methods, we can recover a matrix structure amenable to incremental factorization. We propose a framework ICS that combines a primal-dual method like the Augmented Lagrangian with an incremental Gauss Newton approach that reuses previously computed matrix factorizations. We evaluate ICS on a set of simulated and real-world problems involving equality constraints like object contact and inequality constraints like collision avoidance.

There is a growing interest on formal methods-based robotic planning for temporal logic objectives. In this work, we extend the scope of existing synthesis methods to hyper-temporal logics. We are motivated by the fact that important planning objectives, such as optimality, robustness, and privacy, (maybe implicitly) involve the interrelation between multiple paths. Such objectives are thus hyperproperties, and cannot be expressed with usual temporal logics like the linear temporal logic (LTL). We show that such hyperproperties can be expressed by HyperLTL, an extension of LTL to multiple paths. To handle the complexity of planning with HyperLTL specifications, we introduce a symbolic approach for synthesizing planning strategies on discrete transition systems. Our planning method is evaluated on several case studies.

This paper presents a novel model-free reinforcement learning (RL) framework to design feedback control policies for 3D bipedal walking. Existing RL algorithms are often trained in an end-to-end manner or rely on prior knowledge of some reference joint trajectories. Different from these studies, we propose a novel policy structure that appropriately incorporates physical insights gained from the hybrid nature of the walking dynamics and the well-established hybrid zero dynamics approach for 3D bipedal walking. As a result, the overall RL framework has several key advantages, including lightweight network structure, short training time, and less dependence on prior knowledge. We demonstrate the effectiveness of the proposed method on Cassie, a challenging 3D bipedal robot. The proposed solution produces stable limit walking cycles that can track various walking speed in different directions. Surprisingly, without specifically trained with disturbances to achieve robustness, it also performs robustly against various adversarial forces applied to the torso towards both the forward and the backward directions.

Robots require a detailed understanding of the 3D structure of the environment for autonomous navigation and path planning. A popular approach is to represent the environment using metric, dense 3D maps such as 3D occupancy grids. However, in large environments the computational power required for most state-of-the-art 3D dense mapping systems is compromising precision and real-time capability. In this work, we propose a novel mapping method that is able to build and maintain 3D dense representations for large indoor environments using standard CPUs. Topological global representations and 3D dense submaps are maintained as hybrid global map. Submaps are generated for every new visited place. A place (room) is identified as an isolated part of the environment connected to other parts through transit areas (doors). This semantic partitioning of the environment allows for a more efficient mapping and path-planning. We also propose a method for autonomous exploration that directly builds the hybrid representation in real time.We validate the real-time performance of our hybrid system on simulated and real environments regarding mapping and path-planning. The improvement in execution time and memory requirements upholds the contribution of the proposed work.

This paper proposes a hybrid localization method that fuses Monte Carlo localization (MCL) and convolutional neural network (CNN)-based end-to-end (E2E) localization. MCL is based on particle filter and requires proposal distributions to sample the particles. The proposal distribution is generally predicted using a motion model. However, because the motion model cannot handle unanticipated errors, the predicted distribution is sometimes inaccurate. The use of other ideal proposal distributions, such as the measurement model, can improve robustness against such unanticipated errors. This technique is called importance sampling (IS). However, it is difficult to sample the particles from such ideal distributions because they are not represented in the closed form. Recent works have proved that CNNs with dropout layers represent the posterior distributions over their outputs conditioned on the inputs and the CNN predictions are equivalent to sampling the outputs from the posterior. Therefore, the proposed method utilizes a CNN to sample the particles and fuses them with MCL via IS. Consequently, the advantages of both MCL and E2E localization can be simultaneously leveraged while preventing their disadvantages. Experiments demonstrate that the proposed method can smoothly estimate the robot pose, similar to the model-based method, and quickly re-localize it from the failures, similar to the learning-based method.