IROS 2019 Papers

During the past few years, probabilistic approaches to imitation learning have earned a relevant place in the robotics literature. One of their most prominent features is that, in addition to extracting a mean trajectory from task demonstrations, they provide a variance estimation. The intuitive meaning of this variance, however, changes across different techniques, indicating either variability or uncertainty. In this paper we leverage kernelized movement primitives (KMP) to provide a new perspective on imitation learning by predicting variability, correlations and uncertainty using a single model. This rich set of information is used in combination with the fusion of optimal controllers to learn robot actions from data, with two main advantages: i) robots become safe when uncertain about their actions and ii) they are able to leverage partial demonstrations, given as elementary sub-tasks, to optimally perform a higher level, more complex task. We showcase our approach in a painting task, where a human user and a KUKA robot collaborate to paint a wooden board. The task is divided into two sub-tasks and we show that the robot becomes compliant (hence safe) outside the training regions and executes the two sub-tasks with optimal gains otherwise.

We consider the problem of minimizing the time to approach and land near a target radio beacon at an unknown location with an Unmanned Aerial Vehicle (UAV). We show that a cone-like region exists above the target inside of which bearing measurements of a directional antenna lose directionality: signal recordings in all directions yield similar signal strength. We present a geometric model of this region based on antenna simulations and data collected with a real system. Our main contribution is a strategy that takes advantage of a UAV's ability to change altitude and exploits a special structure occurring when approaching the target beacon from above to reduce the flight time required to land near the beacon. We analyze the performance of our strategy and demonstrate through simulations that by exploiting this structure we can achieve shorter flight times than our previous work.

In this paper, we propose a two-view fusion based convolutional neural network to estimate road areas in urban environments with LiDAR point clouds as input only. The proposed network takes two transformed LiDAR data representations, the LiDAR imageries and the camera-perspective maps, as inputs. It outputs pixel-wise road detection results in both the LiDAR's imagery view and the camera's perspective view simultaneously, in an end-to-end manner. To make better use of the data associations between two representations, we construct a novel mapping layer to transform features from the LiDAR's imagery view to the camera's perspective view in order to strengthen the road detection performance in the camera's perspective view. Experiments on the KITTI-Road dataset show that the proposed network can achieve the state-of-the-art performance among all LiDAR-only methods in real time.

This paper discusses a new teleoperation approach for robotic-assisted tele-echography. A teleoperation architecture in the joint space is presented, taking advantage of kinematic similarity between master and slave manipulators. Haptic force feedback is provided based on the slave control command torque, without using force/torque sensing data. The slave manipulator is controlled in the joint space using computed torque techniques and featuring Kalman Active Observers (AOBs). Experiments in a typical telemedice scenario, assess and validate the teleoperation architecture in a clinical context, with a radiologist performing a robotic-assisted abdominal ultrasound examination on a healthy volunteer.

This paper presents a novel distributed mapping algorithm for multiple resource-constrained robots operating in a rectilinear 2D environment. The algorithm is built upon the sparse wireframe map representation and updating framework in [1]. We propose an exploration strategy based on the labeling of the vertices in the wireframe map, combined with a map-merging interrupt routine that is activated when robots enter into communication range with one another. The maps are not naively merged, but instead the receiving robot verifies the received information before it is assimilated by attempting to drive to the location where the other robot was when communication was established. The robots do not share a global coordinate frame, so prior to a merge the relative map alignment is determined. This is achieved using the random sample consensus (RANSAC) framework with a custom feature which leverages the structure inherent in the wireframe map representation. This results in a lower rate of false-positive matches compared to another state-of-the-art feature used in point cloud alignment, the 4-point congruent set (4PCS). We show our feature to be more robust to false-positive alignments, a common occurrence when attempting to align sparse structures such as wireframe maps. We present high fidelity simulation results in a ROS-Gazebo environment with lidar-equipped TurtleBots1 to highlight the benefits of our algorithm.1The TurtleBot3 burger configuration www.turtlebot.com.

Peg-in-hole is a fine-level manipulation task that requires highly precise location and the normal direction of the hole, which is beyond state-of-the-art object detectors' capability. Therefore, we propose a novel method that trains a robot arm to progressively search for the right inserting pose through trials with both force feedback and visual inputs. Under a reinforcement learning (RL) framework, an agent trying to minimize the number of trials is learned based on the sequentially estimated relative poses with regard to the correct inserting pose. Moreover, our learned dynamics model is transferable. Thanks to our context-independent force and visual feature design, our pre-trained model can be finetuned efficiently for another unseen peg-in-hole case. Extensive experiments show the effectiveness of the proposed framework.

Due to the complexity of modeling the elastic properties of materials, the use of machine learning algorithms is continuously increasing for tactile sensing applications. Recent advances in deep neural networks applied to computer vision make vision-based tactile sensors very appealing for their high-resolution and low cost. A soft optical tactile sensor that is scalable to large surfaces with arbitrary shape is discussed in this paper. A supervised learning algorithm trains a model that is able to reconstruct the normal force distribution on the sensor's surface, purely from the images recorded by an internal camera. In order to reduce the training times and the need for large datasets, a calibration procedure is proposed to transfer the acquired knowledge across multiple sensors while maintaining satisfactory performance.

Planning flight trajectories is important for practical application of flying systems. This topic has been well studied for fixed and rotary winged aerial vehicles, but far fewer works have explored it for flapping systems. Bat Bot (B2) is a bio-inspired flying robot that mimics bat flight, and it possesses the ability to follow a designed trajectory with its on-board electronics and sensing. However, B2's periodic flapping and its complex aerodynamics present major challenges in modeling and planning feasible flight paths. In this paper, we present a generalized approach that uses a model with direct collocation methods to plan dynamically feasible flight maneuvers. The model is made to be both accurate through collection of load cell force data for parameter selection and computationally inexpensive such that it can be used efficiently in a nonlinear solver. We compute the trajectory of launching B2 to a desired altitude and a banked turn maneuver, and we validate our methods with experimental flight results of tracking the launch trajectory with a PD controller.

In this paper, we propose a reinforcement learning-based algorithm for trajectory optimization for constrained dynamical systems. This problem is motivated by the fact that for most robotic systems, the dynamics may not always be known. Generating smooth, dynamically feasible trajectories could be difficult for such systems. Using sampling-based algorithms for motion planning may result in trajectories that are prone to undesirable control jumps. However, they can usually provide a good reference trajectory which a model-free reinforcement learning algorithm can then exploit by limiting the search domain and quickly finding a dynamically smooth trajectory. We use this idea to train a reinforcement learning agent to learn a dynamically smooth trajectory in a curriculum learning setting. Furthermore, for generalization, we parameterize the policies with goal locations, so that the agent can be trained for multiple goals simultaneously. We show result in both simulated environments as well as real experiments, for a 6-DoF manipulator arm operated in position-controlled mode to validate the proposed idea. We compare the proposed ideas against a PID controller which is used to track a designed trajectory in configuration space. Our experiments show that our RL agent trained with a reference path outperformed a model-free PID controller of the type commonly used on many robotic platforms for trajectory tracking.

The steadily increasing popularity of Unmanned Aerial Vehicles (UAVs) is creating new opportunities in diverse fields of technology and business. However, this increase of popularity also raises safety concerns. To tackle the primary concern of keeping the UAV inside a designated region, a novel trajectory estimation algorithm for geo-fencing applications is proposed. We derive the Beta-Trajectory that takes into account constraints in curvature as well as constraints in the change of curvature which is bounded by the maximum roll-rate of the aircraft. We incorporate the Beta-Trajectory into a geo-fencing algorithm. By using our open-source uavAP autopilot, the applicability and necessity of accurate trajectory estimation algorithms for geo-fencing applications are shown on small fixed-wing aircraft. The model and algorithm are validated in high-fidelity simulations as well as in real flight testing.

Recent breakthroughs in the reinforcement learning (RL) community have made significant advances towards learning and deploying policies on real world robotic systems. However, even with current state-of-the-art algorithms and computational resources, these methods are still plagued with high sample complexity, and thus long training times, especially for high degree of freedom (DOF) systems. There are also concerns arising from lack of perceived stability or robustness from emerging policies. This paper aims at mitigating these drawbacks by: (1) modeling a complex, high DOF system with a representative simple one, (2) making explicit use of forward and inverse kinematics without forcing the RL algorithm to “learn” them on its own, and (3) learning locomotion policies in Cartesian space instead of joint space. In this paper, these methods are applied to JPL's Robosimian, but they can be readily used on any system with a base and end effector(s). These locomotion policies can be produced in just a few minutes, trained on a single laptop. We compare the robustness of the resulting learned policies to those of other control methods.

In this paper, a new adaptive tracking controller with learning ability is proposed for fully-constrained cable-driven parallel robots (CDPRs). For these systems, the necessity of maintaining positive and bounded tensions in all cables while coping with disturbances represents a critical control requirement. To achieve this goal, we propose a control law based on adaptive dynamic programming (ADP), with an actorcritic structure. In the critic part, an artificial neural network (NN) approximates the value function which is to evaluate the system performance; in the action part, the controller’s parameters are tuned online to achieve optimal control performance. Additionally, the anti-windup (AW) technique is combined with the adaptive controller to cope with the input saturation problem. The stability of the closed-loop system with the proposed control algorithm is proved using the Lyapunov method. Numerical simulations show the effectiveness of the proposed controller.

Hydrocephalus in the pediatric population is often treated with endoscopic procedures, in which a rigid endoscope provides a working channel and needed visualization for brain manipulation. The lack of flexible and steerable endoscopic tools limits the ability of the surgeon to conduct complex operations. In this paper, we propose the design of a novel tool for such procedures that uses the compliance of a bending flexural joint to build a compact steerable multi-joint tool. These joints are machined into a tube made of a super-elastic material to create two joints near the tip of the tool. The directional stiffness properties of the flexural joint are exploited to minimize inter-joint coupling. We perform a static analysis of each joint and a kinematic analysis of the entire robot. We follow this with the design of a hand-held controller for this robot, and analyze its ability to control multiple degrees-of-freedom of the robot while minimizing the coupling between joints.

Micro autonomous underwater vehicles (μAUVs) allow early-stage experimental testing even in small tanks. In contrast to large facilities, these tanks are usually not equipped with high-fidelity motion capture systems due to high cost and reduced required accuracy for proof-of-concept testing.In this work, we introduce low-cost and open-source motion capture architecture based on fiducial markers for small research tanks. We propose a highly modular approach which allows straight-forward adaptation to individual user needs. We demonstrate the performance of our architecture in two experimental setups.

The use of drones for aerial cinematography has revolutionized several applications and industries that require live and dynamic camera viewpoints such as entertainment, sports, and security. However, safely controlling a drone while filming a moving target usually requires multiple expert human operators; hence the need for an autonomous cinematographer. Current approaches have severe real-life limitations such as requiring fully scripted scenes, high-precision motion-capture systems or GPS tags to localize targets, and prior maps of the environment to avoid obstacles and plan for occlusion.In this work, we overcome such limitations and propose a complete system for aerial cinematography that combines: (1) a vision-based algorithm for target localization; (2) a real-time incremental 3D signed-distance map algorithm for occlusion and safety computation; and (3) a real-time camera motion planner that optimizes smoothness, collisions, occlusions and artistic guidelines. We evaluate robustness and real-time performance in series of field experiments and simulations by tracking dynamic targets moving through unknown, unstructured environments. Finally, we verify that despite removing previous limitations, our system achieves state-of-the-art performance.

The ability to acquire knowledge incrementally and after deployment is of utmost importance for robots operating in the real world. Moreover, robots that have to operate alongside people need to be able to interact in a way that is intuitive for the users, e.g., by understanding and producing natural language. In this paper we present a first prototype of a robot architecture developed for situated lifelong object learning. The system is able to communicate with its users through natural language and perform object learning and recognition on the spot through situated interactions. In this first stage, we evaluate the system in terms of recognition accuracy which gives an indirect measure of the quality of the collected data with the proposed pipeline. Our results show that the robot can use this data for both learning and recognition with acceptable incremental performance. We also discuss limitations and steps that are necessary in order to improve performance as well as to shed some light on system usability.

Imitation of the upper body motions of human demonstrators or animation characters to human-shaped robots is studied in this paper. We present a pipeline for motion retargeting by transferring the joints of interest (JOI) of source motions to the target humanoid robot. To this end, we deploy an optimization-based motion retargeting method utilizing link length modifications of the source skeleton and a task (Cartesian) space fine-tuning of JOI motion descriptors. To evaluate the effectiveness of the proposed pipeline, we use two different 3-D motion datasets from three human demonstrators and an Ogre animation character, Bork, and successfully transfer the motions to four different humanoid robots: DARwIn-OP, COmpliant HuMANoid Platform (COMAN), THORMANG, and Atlas. Furthermore, COMAN and THORMANG are actually controlled to show that the proposed method can be deployed to physical robots.

In this paper we present an initial approach towards reversible robot movement primitives. Our approach is a modification of Dynamic Movement Primitives (DMPs), a widely used framework for robot learning from demonstration. DMPs are based on dynamical systems to guarantee properties such as convergence to a goal state, robustness to perturbation, and the ability to generalize to other goal states. Yet a main limitation of their original formulation is that they do not allow for movements to be reversed. Thus, to execute the same task forwards and backwards would mean to learn two separate primitives. We propose to replace the transformation system in DMPs with the Logistic Differential Equation (LDE), a known time-reversible non-linear system. Similarly to the original DMP formulation, our system's temporal evolution is controlled by a phase system, which in our case is derived from the LDE to guarantee reversibility. We evaluate our approach experimentally with demonstration data from a real robot assembly task, and show comparable properties to those of the original DMP system.

This paper presents a novel system that enables intelligent robots to exhibit realistic body gestures while communicating with humans. The proposed system consists of a listening model and a speaking model used in corresponding conversational phases. Both models are adapted from the sequence-to-sequence (seq2seq) architecture to synthesize body gestures represented by the movements of twelve upper-body keypoints. All the extracted 2D keypoints are firstly 3D-transformed, then rotated and normalized to discard irrelevant information. Substantial videos of human conversations from Youtube are collected and preprocessed to train the listening and speaking models separately, after which the two models are evaluated using metrics of mean squared error (MSE) and cosine similarity on the test dataset. The tuned system is implemented to drive a virtual avatar as well as Pepper, a physical humanoid robot, to demonstrate the improvement on conversational interaction abilities of our method in practice.