ICRA 2017 Papers

With systems performing Simultaneous Localization And Mapping (SLAM) from a single robot reaching considerable maturity, the possibility of employing a team of robots to collaboratively perform a task has been attracting increasing interest. Promising great impact in a plethora of tasks ranging from industrial inspection to digitization of archaeological structures, collaborative scene perception and mapping are key in efficient and effective estimation. In this paper, we propose a novel, centralized architecture for collaborative monocular SLAM employing multiple small Unmanned Aerial Vehicles (UAVs) to act as agents. Each agent is able to independently explore the environment running limited-memory SLAM onboard, while sending all collected information to a central server, a ground station with increased computational resources. The server manages the maps of all agents, triggering loop closure, map fusion, optimization and distribution of information back to the agents. This allows an agent to incorporate observations from others in its SLAM estimates on the fly. We put the proposed framework to the test employing a nominal keyframe-based monocular SLAM algorithm, demonstrating the applicability of this system in multi-UAV scenarios.

This paper presents a motion planning approach for cooperative transportation using aerial robots. We describe a framework based on Parametric Dynamic Movement Primitives (PDMPs) for coordinating multiple aerial robots and their manipulators quickly in an environment cluttered with obstacles. In order to emulate the optimal motion, we combine PDMPs and Rapidly Exploring Randomized Trees star (RRT*) by using the results of RRT* as demonstrations for PDMPs. For efficient description of the motions corresponding to the environment, we utilize Gaussian Process Regression (GPR) to acquire of the explicit relationship between environmental parameters and style parameters of PDMPs which decide the motions. Simulation and experiment results are attached to validate the proposed framework.

Motion planning as trajectory optimization requires generating trajectories that minimize a desired objective function or performance metric. Finding a globally optimal solution is often intractable in practice: despite the existence of fast motion planning algorithms, most are prone to local minima, which may require re-solving the problem multiple times with different initializations. In this work we provide a novel motion planning algorithm, GPMP-GRAPH, that considers a graph-based initialization that simultaneously explores multiple homotopy classes, helping to contend with the local minima problem. Drawing on previous work to represent continuous-time trajectories as samples from a Gaussian process (GP) and formulating the motion planning problem as inference on a factor graph, we construct a graph of interconnected states such that each path through the graph is a valid trajectory and efficient inference can be performed on the collective factor graph. We perform a variety of benchmarks and show that our approach allows the evaluation of an exponential number of trajectories within a fraction of the computational time required to evaluate them one at a time, yielding a more thorough exploration of the solution space and a higher success rate.

A novel algorithm for planning robotic manipulation tasks is presented in which the base position and joint motions of a robot are simultaneously optimized to follow a smooth desired end-effector trajectory. During the optimization routine, the manipulator's base position and joint motions are planned simultaneously by strategically moving a set of virtual robot arms (each representing a single configuration in a sequence) branching from a common base to a number of assigned target poses associated with a task. Additional goals (e.g. collision avoidance) and hard constraints, including joint limits are also incorporated. The optimization problem at the core of this method is a quadratic program, allowing constrained high-dimensional problems to be solved in very little time. This method has successfully planned motions allowing an 8-DOF manipulator to paint walls, and has proven to be highly efficient and scalable in practice.

This paper presents a new approach for active tactile sensation that utilizes soft morphological computation. This work is inspired by human finger's wet-induced wrinkles, which appear after a long time soaking in water, and has been indicated as an efficient means for enhancement of gripping in wet environment. We created a tactile sensing system that is an integration of actuation (pneumatic actuator) and sensing elements (strain gauges). This device can change its morphology so that the posture of embedded sensing elements can vary, then generate different responses depending on the sensing tasks. As a result, this device can actively select its sensing functions depending on different sensing tasks. In this paper, the sensing device is both sensitive to indentation contact and sliding action on its surface by using only one type of strain gauge. This preliminary work is an example of soft morphological control in sensing, and expected to open a new trend in development of tactile sensing system.

Real-time monocular SLAM is increasingly mature and entering commercial products. However, there is a divide between two techniques providing similar performance. Despite the rise of ‘dense’ and ‘semi-dense’ methods which use large proportions of the pixels in a video stream to estimate motion and structure via alternating estimation, they have not eradicated feature-based methods which use a significantly smaller amount of image information from keypoints and retain a more rigorous joint estimation framework. Dense methods provide more complete scene information, but in this paper we focus on how the amount of information and different optimisation methods affect the accuracy of local motion estimation (monocular visual odometry). This topic becomes particularly relevant after the recent results from a direct sparse system. We propose a new method for fairly comparing the accuracy of SLAM frontends in a common setting. We suggest computational cost models for an overall comparison which indicates that there is relative parity between the approaches at the settings allowed by current serial processors when evaluated under equal conditions.

We present an approach that allows the Georgia Tech Miniature Autonomous Blimp (GT-MAB) to detect and follow a human. This accomplishment is the first Human Robot Interaction (HRI) demonstration between an uninstrumented human and a robotic blimp. GT-MAB is an ideal platform for HRI missions because it is safe to humans and can support sufficient flight time for HRI experiments. However, due to complex aerodynamic influence on the blimp, the human following task for GT-MAB with a single on-board camera is a challenging problem. We integrate Haar face detector and KLT feature tracker to achieve robust human tracking. After a human face is detected in the real-time video stream, we estimated the 3D positions of the human with respect to GT-MAB. Visionbased PID controllers are designed based on estimated relative position and the motion primitives of GT-MAB such that it can achieve stable and continuous human following behavior. Experimental results are presented to demonstrate the human following capability on GT-MAB.

RGBD SLAM systems have shown impressive results, but the limited field of view (FOV) and depth range of typical RGBD cameras still cause problems for registering distant frames. Monocular SLAM systems, in contrast, can exploit wide-angle cameras and do not have the depth range limitation, but are unstable for textureless scenes. We present a SLAM system that uses both an RGBD camera and a wide-angle monocular camera for combining the advantages of the two sensors. Our system extracts 3D point features from RGBD frames and 2D point features from monocular frames, which are used to perform both RGBD-to-RGBD and RGBD-to-monocular registration. To compensate for different FOV and resolution of the cameras, we generate multiple virtual images for each wide-angle monocular image and use the feature descriptors computed on the virtual images to perform the RGBD-to-monocular matches. To compute the poses of the frames, we construct a graph where nodes represent RGBD and monocular frames and edges denote the pairwise registration results between the nodes. We compute the global poses of the nodes by first finding the minimum spanning trees (MSTs) of the graph and then pruning edges that have inconsistent poses due to possible mismatches using the MST result. We finally run bundle adjustment on the graph using all the consistent edges. Experimental results show that our system registers a larger number of frames than using only an RGBD camera, leading to larger-scale 3D reconstruction.

A self-reconfigurable modular robotic system has potential to carry out a wide variety of tasks in various fields and conditions. For realizing such versatileness and robustness, the compositional modules have to be designed so that the combined structure is equipped with high degree of freedom of the self-reconfiguration. We propose a new design of a modular robot using helically magnetized axes. The axes are arranged at all of the edges of the cubic-shaped module and contribute to both connection and transformation among the modules. The mechanism enables the modules to construct a 3-D orthogonal lattice structure, execute slide movement to neighboring vacant grid positions, and change the structural configuration by themselves. This paper firstly describes the conceptual design and fabrication method of the helically magnetized axis. Then, total design and specification of the module are explained. Finally, experimental results to evaluate the performance of connection and transformation are presented.

This paper presents a modified nonlinear pneumatic model for pneumatic force servo systems. The modified model is proposed in order to estimate pressures accurately for both chambers of a pneumatic cylinder by adopting flow coefficient maps, which is different from a conventional model whose flow coefficient is constant. The simulated data from the model is also compared to the experiment result and shows its accuracy compared to a widely used conventional model. The modified model and the experimental procedure are described and verified in this paper as well. Finally, the applicability of the proposed model to model-based controller design is also shown through the experiment of force servo controls based on a force sensor.

This paper addresses a difficulty in large-scale long term laser localisation - how to deal with scene change. We pose this as a distraction suppression problem. Urban driving environments are frequently subject to large dynamic outliers, such as buses, trucks etc. These objects can mask the static elements of the prior map that we rely on for localisation. At the same time some objects change shape in a way that is less dramatic but equally pernicious during localisation - for example trees over seasons and in wind, shop fronts and doorways. In this paper, we show how we can learn in high resolution, the areas of our map that are subject to such distractions (low value data) in a place-dependent approach. We demonstrate how to utilise this model to select individual laser measurements for localisation. Specifically, by leveraging repeated operation over weeks and months, for each point in our map pointcloud we build distributions of the errors associated with that point for multiple localisation passes. These distributions are then used to determine the legitimacy of laser measurements prior to their use in localisation. We demonstrate distraction suppression as a front-end process to large scale localiser by incrementally adding 50km of error data to our base map and show that robustness is improved over the base system with a further 10km of urban driving.

In shared autonomy, a robot and human user both have some level of control in order to achieve a shared goal. Choosing the balance of control given to the user and the robot can be a challenging problem since different users have different preferences and vary in skill levels when operating a robot. We propose using a novel formulation of Partially Observable Markov Decision Process (POMDP) to represent a model of the user's expertise in controlling the robot. The POMDP uses observations from the user's actions and from the environment to update the belief of the user's skill and chooses a level of control between the robot and the user. The level of control given between the user and the robot is encapsulated in macro-action controllers. A user study was run to test the performance of our formulation. Users drive a simulated robot through an obstacle-filled map while the POMDP model chooses appropriate macro-action controllers based on the belief state of the user's skill level. The results of the user study show that our model can encapsulate user skill. The results also show that using the controller with greater robot autonomy helped users of low skill avoid obstacles more than it helped users of high skill.

Parallel continuum robots consist of a parallel arrangement of flexible legs and are dexterous, compliant, and easily miniaturized for minimally invasive surgery. By design, parallel continuum robots exhibit large, nonlinear deformations in their legs to achieve multi-DOF end effector articulation, but excess leg bowing can limit their reachable workspace, especially for long slender designs. In this paper, we investigate a parallel continuum robot design with a passive spring backbone carrying disks that constrain the legs at intermediate points. The constraints route the legs in helical paths around the backbone and prevent large divergence of the legs, expanding the reachable workspace for slender form factors while preserving the manipulator's six degrees of freedom. We present a novel forward and inverse kinematics model, based on Cosserat rod theory, that accommodates general leg routing paths and any number of intermediate constraint disks. We also explore manipulator workspace with experiments and simulations, demonstrating that intermediate constraints expand the reachable workspace of slender parallel continuum robots.

The modeling and identification of a mechanical system is the most important issue for many control systems in order to realize the desired control specifications. In particular, the friction characteristics often deteriorate the control performance, such as in the fast and precise positioning performance in industrial robots, the force estimation accuracy based on a disturbance observer, and the posture control performance of an inverted pendulum robot. Rolling friction tends to cause overshoot, undershoot, or limit cycles of the target value in positioning systems. In previous research, some model structures for rolling friction have been proposed to express the hysteresis characteristics in order to overcome these control issues. However, it is difficult to identify the correct parameters for precise modeling. In this paper, the modeling of rolling friction based on a Recurrent Neural Network (RNN) using Long Short-Term Memory (LSTM) is proposed to precisely express the rolling friction characteristics. The initial value design of the RNN during supervised learning is also presented to achieve a better model. The effectiveness of the proposed approach is verified by comparison with conventional friction models using an actual experimental setup.

For robots to be a part of our daily life, they need to be able to navigate among crowds not only safely but also in a socially compliant fashion. This is a challenging problem because humans tend to navigate by implicitly cooperating with one another to avoid collisions, while heading toward their respective destinations. Previous approaches have used handcrafted functions based on proximity to model human-human and human-robot interactions. However, these approaches can only model simple interactions and fail to generalize for complex crowded settings. In this paper, we develop an approach that models the joint distribution over future trajectories of all interacting agents in the crowd, through a local interaction model that we train using real human trajectory data. The interaction model infers the velocity of each agent based on the spatial orientation of other agents in his vicinity. During prediction, our approach infers the goal of the agent from its past trajectory and uses the learned model to predict its future trajectory. We demonstrate the performance of our method against a state-of-the-art approach on a public dataset and show that our model outperforms when predicting future trajectories for longer horizons.

The concept of utilizing the Coandia effect to produce stable flight in saucer type Aerial Vehicles (AV) was first conceived in 1935 by Henri Marie Coanda. Coanda's proposed AV design remained a relatively unexplored curiosity for decades until the recent surge in popularity of small Unmanned Aerial Vehicles (UAV) revitalized interest in the topic. Coandä effect UAV offer some distinct advantages over standard multirotors due to their robust frame design, which offers a unique mix of crash resistance, flight safety, and flight performance making them ideal for Human-Robot Interaction (HRI) applications. While much work has been performed in the area of characterization for Coanda surfaces and improved mechanical design of saucer type UAV, little work has been found in the literature discussing stability and attitude control of the platform. This work seeks to contribute to the study of these drones by introducing an approximate servo mapping for desired control effort of a particular prototype, and an experimental control law implemented within the framework of a modified commercially available flight stack.

Robotic camera systems have recently drawn attention in minimally invasive surgeries(MIS). Control and manipulation of these systems during traversing abdominal cavity is associated with camera-tissue interaction. This paper demonstrates a theoretical and experimental analysis of a wireless laparoscopic camera's interaction with abdominal wall during MIS. A mechanical model is developed to represent the behavior of abdominal wall bulk tissue by considering skin, fat, muscle and connective tissues layers which predicts behavior of the tissue during interaction with laparoscopic camera. The model was implemented in ABAQUS to analyze the camera-tissue interaction and find interaction forces generated during contact and motion with different linear and rotational speeds. Simulations were validated by experiments on porcine tissue which can be used for proper control of insertable camera. A noninvasive method is proposed to measure the mechanical properties of each patient's abdominal wall tissue at start of MIS in order to tune the control system to optimize the interaction depth and opposite forces. This method features preventing overload damage and camera fall during MIS.

To many unmanned aerial vehicle (UAV) designs, the lack of information about the wind speed and direction is a limiting factor in achieving robust outdoor flight. This paper addresses the problem of wind estimation onboard a hovering vertical take-off and landing (VTOL) tailsitter UAV. The proposed estimation framework makes use of the standard onboard sensor suite: inertial measurement unit (IMU), global positioning system (GPS) and a magnetometer. No additional airspeed sensor is needed. As a result, the autopilot is provided with an estimate of the wind velocity vector in the horizontal (north-east) plane. An aerodynamic model of the vehicle has been derived and used in a Kalman filter framework to estimate the horizontal wind velocity vector in real-time. The wind estimator has been implemented onboard the UAVs autopilot and validated in real flight. As a result, we successfully obtain the direction and speed of the wind with an estimation accuracy close to the accuracy range of the ground truth measurement. Furthermore, the derived grey-box model allows to generalise the framework to different airframes.

This paper addresses the problem of trajectory optimization for the transition of a Vertical Take-off and Landing (VTOL) tailsitter Unmanned Aerial Vehicle (UAV). The proposed strategy performs a model based optimization, where the model represents the closed-loop dynamics of the UAV with low-level control, ensuring attitude stabilization over the whole trajectory. We discuss the design of an optimization framework, vehicle modeling, and elaborate on the cost function construction. An additional feedback gain is implemented on the throttle channel with altitude discrepancies as its input to provide some level of robustness to wind disturbances. The overall approach is verified in simulation and experimental results with a focus on optimization of the back-transition (cruise-to-hover) of the Wingtra S100 VTOL tailsitter.