IROS 2019 Papers

We propose a novel pose estimation method for geometric vision of omni-directional cameras. On the basis of the regularity of the pixel movement after camera pose changes, we formulate and prove the sinusoidal relationship between pixels movement and camera motion. We use the improved Fourier-Mellin invariant (iFMI) algorithm to find the motion of pixels, which was shown to be more accurate and robust than the feature-based methods. While iFMI works only on pin-hole model images and estimates 4 parameters (x, y, yaw, scaling), our method works on panoramic images and estimates the full 6 DoF3D transform, up to an unknown scale factor. For that we fit the motion of the pixels in the panoramic images, as determined by iFMI, to two sinusoidal functions. The offsets, amplitudes and phase-shifts of the two functions then represent the 3D rotation and translation of the camera between the two images. We perform experiments for 3D rotation, which show that our algorithm outperforms the feature-based methods in accuracy and robustness. We leave the more complex 3D translation experiments for future work.

Multiagent reinforcement learning (MARL) algorithms have been demonstrated on complex tasks that require the coordination of a team of multiple agents to complete. Existing works have focused on sharing information between agents via centralized critics to stabilize learning or through communication to improve performance, but do not generally consider how information can be shared between agents to address the curse of dimensionality in MARL. We posit that a multiagent problem can be decomposed into a multi-task problem where each agent explores a subset of the state space instead of exploring the entire state space. This paper introduces a multiagent actor-critic algorithm for combining knowledge from homogeneous agents through distillation and value-matching that outperforms policy distillation alone and allows further learning in discrete and continuous action spaces.

Accurate endoscope localization is a key element for Computer-Assisted Laparoscopic Surgery (CALS). External localizers track the endoscope pose and reconstruct the camera geometry through a calibration matrix. Perspective-n-Points (PnP) methods estimate this matrix from a set of paired point coordinates, given by the simultaneous position of a spherical marker on the image (2D coordinates) and in the external localizer frame (3D coordinates). In this paper, the PnP calibration is restated as an iterative Perspective-n-Spheres (PnS) problem: the spherical geometry of the marker is taken into account when computing the projection of its 3D center into the 2D image plane. Using calibration residuals as evaluation matrix, the proposed PnS provides better results compared to the State-of-the-Art (SoA) PnP methods. Furthermore, results show that the calibration yields different estimates due to different camera poses in the working volume. This is assumed to result from a lack of absolute precision in the marker localization across the external localizer workspace.

Swarm robots operate as autonomous agents and a swarm as a whole gets autonomous by its capability of collective decision-making. Despite intensive research on models of collective decision-making, the implementation in multi-robot systems is still challenging. Here, we advance the state of the art by introducing more plasticity to the decision-making process and by increasing the scenario difficulty. Most studies on large-scale multi-robot decision-making are limited to one instance of an iterated exploration-dissemination phase followed by successful and permanent convergence. We investigate a dynamic environment that requires constant collective monitoring of option qualities. Once a significant change in qualities is detected by the swarm, it has to collectively reconsider its previous decision accordingly. This is only possible by preventing lock-ins, a global consensus state of no return (i.e., a dominant majority of robots prevents the swarm from switching to another, possibly better option). In addition, we introduce a scenario of increased difficulty as the robots must locate themselves to assess the quality of an option. Using local communication, swarm robots propagate hop-count information throughout the swarm to form a global reference frame. We successfully validate our implementation in many swarm robot experiments concerning robustness to disruptions of the reference frame, scalability, and adaptivity to a dynamic environment.

We present the Goal Uncertain Stochastic Shortest Path (GUSSP) problem - a general framework to model path planning and decision making in stochastic environments with goal uncertainty. The framework extends the stochastic shortest path (SSP) model to dynamic environments in which it is impossible to determine the exact goal states ahead of plan execution. GUSSPs introduce flexibility in goal specification by allowing a belief over possible goal configurations. The unique observations at potential goals helps the agent identify the true goal during plan execution. The partial observability is restricted to goals, facilitating the reduction to an SSP with a modified state space. We formally define a GUSSP and discuss its theoretical properties. We then propose an admissible heuristic that reduces the planning time using FLARES - a start-of-the-art probabilistic planner. We also propose a determinization approach for solving this class of problems. Finally, we present empirical results on a search and rescue mobile robot and three other problem domains in simulation.

When robots perform manipulation tasks, they need to determine their own movement, as well as how to make and break contact with objects in their environment. Reasoning about the motions of robots and objects simultaneously leads to a constrained planning problem in a high-dimensional state-space. Additionally, when environments change dynamically motions must be computed in real-time. To this end, we propose a feedback planner for manipulation. We model manipulation as constrained motion and use this model to automatically derive a set of constraint-based controllers. These controllers are used in a switching-control scheme, where the active controller is chosen by a reinforcement learning agent. Our approach is capable of addressing tasks with second-order dynamics, closed kinematic chains, and time-variant environments. We validated our approach in simulation and on a real, dual-arm robot. Extensive simulation of three distinct robots and tasks show a significant increase in robustness compared to a previous approach.

We present a method that plans motions for a concentric tube robot to automatically reach surgical targets inside the body while avoiding obstacles, where the patient's anatomy is represented by point clouds. Point clouds can be generated intra-operatively via endoscopic instruments, enabling the system to update obstacle representations over time as the patient anatomy changes during surgery. Our new motion planning method uses a combination of sampling-based motion planning methods and local optimization to efficiently handle point cloud data and quickly compute high quality plans. The local optimization step uses an interior point optimization method, ensuring that the computed plan is feasible and avoids obstacles at every iteration. This enables the motion planner to run in an anytime fashion, i.e., the method can be stopped at any time and the best solution found up until that point is returned. We demonstrate the method's efficacy in three anatomical scenarios, including two generated from endoscopic videos of real patient anatomy.

Last-mile delivery systems commonly propose the use of autonomous robotic vehicles to increase scalability and efficiency. The economic inefficiency of collecting accurate prior maps for navigation motivates the use of planning algorithms that operate in unmapped environments. However, these algorithms typically waste time exploring regions that are unlikely to contain the delivery destination. Context is key information about structured environments that could guide exploration toward the unknown goal location, but the abstract idea is difficult to quantify for use in a planning algorithm. Some approaches specifically consider contextual relationships between objects, but would perform poorly in object-sparse environments like outdoors. Recent deep learning-based approaches consider context too generally, making training/transferability difficult. Therefore, this work proposes a novel formulation of utilizing context for planning as an image-to-image translation problem, which is shown to extract terrain context from semantic gridmaps, into a metric that an exploration-based planner can use. The proposed framework has the benefit of training on a static dataset instead of requiring a time-consuming simulator. Across 42 test houses with layouts from satellite images, the trained algorithm enables a robot to reach its goal 189% faster than with a context-unaware planner, and within 63% of the optimal path computed with a prior map. The proposed algorithm is also implemented on a vehicle with a forward-facing camera in a high-fidelity, Unreal simulation of neighborhood houses.

We present a novel method enabling robots to quickly learn to manipulate objects by leveraging a motion planner to generate “expert” training trajectories from a small amount of human-labeled data. In contrast to the traditional sense-plan-act cycle, we propose a deep learning architecture and training regimen called PtPNet that can estimate effective end-effector trajectories for manipulation directly from a single RGB-D image of an object. Additionally, we present a data collection and augmentation pipeline that enables the automatic generation of large numbers (millions) of training image and trajectory examples with almost no human labeling effort.We demonstrate our approach in a non-prehensile tool-based manipulation task, specifically picking up shoes with a hook. In hardware experiments, PtPNet generates motion plans (open-loop trajectories) that reliably (89% success over 189 trials) pick up four very different shoes from a range of positions and orientations, and reliably picks up a shoe it has never seen before. Compared with a traditional sense-plan-act paradigm, our system has the advantages of operating on sparse information (single RGB-D frame), producing high-quality trajectories much faster than the expert planner (300ms versus several seconds), and generalizing effectively to previously unseen shoes. Video available at https://youtu.be/voIkyiBtwn4.

Recent advances in on-policy reinforcement learning (RL) methods enabled learning agents in virtual environments to master complex tasks with high-dimensional and continuous observation and action spaces. However, leveraging this family of algorithms in multi-fingered robotic grasping remains a challenge due to large sim-to-real fidelity gaps and the high sample complexity of on-policy RL algorithms. This work aims to bridge these gaps by first reinforcement-learning a multi-fingered robotic grasping policy in simulation that operates in the pixel space of the input: a single depth image. Using a mapping from pixel space to Cartesian space according to the depth map, this method transfers to the real world with high fidelity and introduces a novel attention mechanism that substantially improves grasp success rate in cluttered environments. Finally, the direct-generative nature of this method allows learning of multi-fingered grasps that have flexible end-effector positions, orientations and rotations, as well as all degrees of freedom of the hand.

In this paper, we introduce a novel variational approach to estimate the scene flow from RGB-D images. We regularize the ill-conditioned problem of scene flow estimation in a unified framework by enforcing piecewise rigid motion through decomposition into rotational and translational motion parts. Our model crucially regularizes these components by an L0 “norm”, thereby facilitating implicit motion segmentation in a joint energy minimization problem. Yet, we also show that this energy can be efficiently minimized by a proximal primal-dual algorithm. By implementing this approximate L0 rigid motion regularization, our scene flow estimation approach implicitly segments the observed scene of into regions of nearly constant rigid motion. We evaluate our joint scene flow and segmentation estimation approach on a variety of test scenarios, with and without ground truth data, and demonstrate that we outperform current scene flow techniques.

Robot teleoperation via motion mapping has been demonstrated to be an efficient and intuitive approach for controlling and teaching the whole-body motion coordination of humanoid robots. However, the physical fatigue in the usage of such robot teleoperation interfaces may prevent this approach to be widely used in large scale by diverse workforce populations. As a result, this paper conducts a user study to investigate the physical fatigue of teleoperators in the whole-body motion mapping teleoperation of a mobile humanoid assistive robot. Through a Vicon motion capture system, participants teleoperated the robot to perform general purpose assistive tasks that involve reaching-to-grasp, bimanual manipulation, loco-manipulation and human-robot interaction. We assess the physical fatigue based on surface electromyography (sEMG) measurement, and compare it between different tasks and muscles. Our analysis results indicate that: (1) Fatigue happens more in the tasks that involve more precise manipulation and steady posture maintenance; (2) Deltoids, Biceps and Trapezius are used more for such tasks and thus have more fatigue than others. These findings imply that automating the fatigue-causing task components may reduce the physical fatigue in motion mapping teleoperation.

Many existing mobile robotic telepresence systems have equipped with two web cameras, one is a forward-facing camera (FF camera) for video communication, and the other is a downward-facing camera (DF camera) for robot navigation. In this paper, we present a new framework of autonomous person-following for telepresence robots using the two web cameras. Based on correlation filters tracking methods, we use the FF camera to track the upper body of a person and the DF camera to localize and track the person's feet. We improve the robustness of feet trackers, consisting of a left foot tracker and a right foot tracker, by making full use of the spatial constraints of the human body parts. We conducted experiments on tracking in different environmental situations and real person-following scenario to evaluate the effectiveness of our method.

This paper presents a novel design of four permanent magnet based actuator for magnetic drug targeting of therapeutic microrobots. Drugs are transported through therapeutic magnetic boluses composed of magnetic particles controlled by magnetic gradients. In this study, to maximize the effect of the treatment and minimize adverse effects on the patient, a magnetic actuator have been developed to wirelessly control microrobots in a fluidic environment using four external magnets. Experimental validation is carried using the robotic arm Fanuc LR Mate 200iD to demonstrate the steerability of the magnetic microrobots under different trajectory constraints in viscous fluidic environment.

For underactuated bipedal robots, it is important to compensate disturbances which affects the zero dynamics to realize a stable locomotion when we use output zeroing controllers. In order to deal with such disturbances, this paper presents a framework of a periodic trajectory planning and a robust output zeroing controller using a disturbance model learned by Gaussian process regression (GPR). In particular, we propose a control method considering the modeling uncertainty of the disturbance model by using a variance information of GPR model. We show the effectiveness of the proposed method through a numerical simulation of walking control of an underactuated robot model affected by an external force.

We present a novel method of using terminal costs in the construction of a receding horizon search path. We prove that the proposed method of constructing search paths provides a theoretical lower bound on the performance of the search path. Our result can be interpreted as ensuring that the receding horizon path performs no worse in expectation than a given sub-optimal search path. This result is especially practical for subsea applications where, due to use of side-scan sonar in search applications, search paths typically consist of parallel straight lines. Thus for subsea search applications, our approach ensures that expected performance is no worse than the usual subsea search path, and it might be much better. We demonstrate the efficacy of the proposed method by planning search paths in simulation using real-world data that was acquired by an autonomous underwater vehicle during a subsea survey of Boston Harbor.

Mobile robots moving in crowded environments have to navigate among pedestrians safely. Ideally, the way the robot avoids the pedestrians should not only be physically safe but also perceived safe and comfortable. Despite the rich literature in collision-free crowd navigation, limited research has been conducted on how humans perceive robot behaviors in the navigation context. In this paper, we implement three local pedestrian avoidance strategies inspired by human avoidance behaviors on a self-balancing mobile robot and evaluate their perception in a human-robot crossing scenario through a large-scale user study with 98 participants. The study reveals that the avoidance strategies positively affect the participants' perception of the robot's safety, comfort, and awareness to different degrees. Furthermore, the participants perceive the robot as more intelligent, friendly and reliable in the last trial than in the first even with the same strategy.

We propose and demonstrate a framework called perception as prediction for autonomous driving that uses general value functions (GVFs) to learn predictions. Perception as prediction learns data-driven predictions relating to the impact of actions on the agent's perception of the world. It also provides a data-driven approach to predict the impact of the anticipated behavior of other agents on the world without explicitly learning their policy or intentions. We demonstrate perception as prediction by learning to predict an agent's front safety and rear safety with GVFs, which encapsulate anticipation of the behavior of the vehicle in front and in the rear, respectively. The safety predictions are learned through random interactions in a simulated environment containing other agents. We show that these predictions can be used to produce similar control behavior to an LQR-based controller in an adaptive cruise control problem as well as provide advanced warning when the vehicle behind is approaching dangerously. The predictions are compact policy-based predictions that support prediction of the long term impact on safety when following a given policy. We analyze two controllers that use the learned predictions in a racing simulator to understand the value of the predictions and demonstrate their use in the real-world on a Clearpath Jackal robot and an autonomous vehicle platform.

For commercially successful ground robots, high degree of autonomy, low manufacturing and maintenance cost, as well as minimized deployment limitations in different environments are essential attributes. To deliver an `anywhere deployable' product, it is impractical to rely on one single sensor or one single piece of algorithm to overcome all related challenges. Instead, the entire robotic system should be dedicated designed, including the choices of sensors, processors, algorithm integration for various functionality, and so on.This paper presents our design of perception system for commercial ground robots, which is able to operate in most common environments. The designed system is equipped with low-cost sensors and processors. The first key contribution of this paper is the design of the robotic sensory system, which includes a monocular camera, a 2D laser range finder (LRF), wheel encoders, and an inertial measurement unit (IMU). Our sensory system can be built at a cost of as low as $100. Furthermore, the selected sensors provide complementary characteristics for perception of both robot ego-motion and its surrounding environments, which are the prerequisites for `anywhere' deployment. The second key contribution of this paper is that a complete set of technologies is proposed based on our sensor systems, including sensor calibration (factory calibration and online calibration), localization (environmental exploring and re-localization), as well as mapping. The proposed methodology includes both efficient engineering implementation and theoretical novelty for high performance systems. Experimental results from our robotic testing platform and off-the-shelf commercial robots are presented. These results demonstrate that the proposed system can be deployed in various environmental conditions without performance compromise.