Previous publications of our institute describe a robust vehicle tracking system for daylight conditions. This paper presents an improved vehicle tracking system which is able to detect and track the convoy leader also by twilight and night. The primary sensor equipment consists of a daytime camera, a LiDAR and an inertial navigation system. An expansion with a thermal and a lowlight camera was necessary to be robust against any illumination conditions. The system is capable of estimating the relative 3D position and orientation, the velocity and the steering angle of a convoy leader precisely in real-time. This makes it possible to follow the convoy leader's track. Another novelty is coupling a Kalman filter with a particle filter for higher stability and accuracy in vehicle tracking. The tracking system shows excellent functionality while driving more than 50km fully autonomously in urban- and unstructured environments at night.
The tilted assumption, a remnant of analog cameras, is still in use. It seeks to compensate the off squareness of the image plane with the lens axis, but in fact creates a systematic shape alteration by introducing a scale variation in the image, and adds shear with the use of a skew parameter. It results in an image center bias which in turn offsets every single parameter estimate in the camera model. We disclose our own exact solution to the internal camera model, modeling the image plane as a pure projection, our related camera calibration method, and discuss the various improvements resulting from our find in almost every performance aspect of digital imaging.
In this paper, we propose a novel approach to control autonomous robots to achieve a desired linear formation during movement towards the target position. Firstly, one robot, which has the closest distance to the target, is selected as the leader of the swarm. The desired formation is built based on the relative position between this leader and the target. Secondly, the trajectory of the remaining robots towards the optimal positions in the desired formation is driven by the artificial force fields. These force fields consist of the local and global attractive potential fields surrounding each virtual node in the desired formation. Furthermore, an orientation controller is added in order to guarantee that the desired formation is always headed in the invariant direction to the target position. In addition, the local repulsive force fields around each robot and obstacle are employed in order to avoid collisions during movement. The stability of a swarm following a desired collinear formation in invariant direction towards the target is verified in simulations and experiments.
This paper presents the overview and the introduction of commercialized robotic products in the industrial service, maintenance and repair sectors. General facts of the industrial service are briefly described, and then we focus on four specific applications including motor /generator inspection, solar panel inspection /cleaning, tank inspection and pipe inspection. For each application, service process characteristics, operational details, technical challenges, requirements are described. Robotics solutions with commercialized products of each application area were introduced and detailed with special features and specification.
This paper proposes a human detection-based cognitive system for robots to work in human-existing environment and keep the safety of humans. An integrated system is implemented with perception, recognition, reasoning, decision-making, and action. Without using any traditional safety cages, a vision-based detection system is implemented for robots to monitor the environment and to detect humans. Subsequently, reasoning and decision making enables robots to evaluate the current safety-related situation for humans and provide corresponding safety signals. The decision making is based on maximizing the productivity of the robot in the manipulation process and keep the safety of humans in the environment. The system is implemented with a Baxter humanoid robot and a PowerBot mobile robot. Practical experiments and simulation experiments are carried out to validate our design.
Today's market for domotic systems is dominated by proprietary communication protocols incompatible with each other. This not only prevents engineers from building highly integrated home automation and domestic robotic systems, but keeps prices artificially high due to a lack of competition in the market since the customers have to restrict their choices to components compatible with the system already deployed. In this paper, we propose a communication protocol for fully integrated domotic systems running on top of house network, protocol which is simple enough to accommodate as peers microcontroller based sensors but scalable enough to handle a house wide distributed computer vision system for domestic robot assisted living. Based on a house centered philosophy, the protocol is putting emphasis on privacy and protection of the residents.
This paper presents two prototype fast walking gaits for the quadruped robot RoboSimian, along with experimental results for each. The first gait uses a statically stable one-at-a-time swing-leg crawl. The second gait uses a two-at-a-time swingleg motion, which requires deliberate planning of zero-moment point (ZMP) to balance the robot on a narrow support base. Of particular focus are the development of practical means to exploit the fact that RoboSimian has high-dimensionality, with seven actuators per limb, as a means of partially overcoming low joint velocity limits at each joint. For both gaits, we use an inverse kinematics (IK) table that has been designed to maximize the reachable workspace of each limb while minimizing joint velocities during end effector motions. Even with the simplification provided by use of IK solutions, there are still a wide range of variables left open in the design of each gait. We discuss these and present practical methodologies for parameterizing and subsequently deriving approximate time-optimal solutions for each gait type, subject to joint velocity limits of the robot and to real-world requirements for safety margins in maintaining adequate balance. Results show that careful choice of parameters for each of the gaits improves their respective walking speeds significantly. Finally, we compare the fastest achievable walking speeds of each gait and find they are nearly equivalent, given current performance limits of the robot.
We propose a new extend function for Rapidly-Exploring Randomized Tree (RRT) algorithms that expands along a curve, obeying velocity and acceleration limits, rather than using straight-line trajectories. This results in smooth, feasible trajectories that can readily be applied in robotics applications. Our main focus is the implementation of such methods on RoboSimian, a quadruped robot competing in the DARPA Robotics Challenge (DRC). Planning in a high-dimensional space is also a large consideration in the evaluation of the techniques discussed in this paper as motion planning for RoboSimian requires a search over a 16-dimensional space. In our experiments, we show that our approach produces results that are comparable to the standard RRT solutions in a two-dimensional space and significantly outperforms the latter in a higher-dimensional setting both in computation time and in algorithm reliability.
Rapid Robot Prototyping (R2P) is an open source HW/SW framework providing components for the rapid development of robotic applications. R2P framework components reuse and easy integration is obtained through an embedded real-time publish/subscribe middleware which allows distributed control loops to be set up in a flexible way. R2P aims at increasing hardware and software reuse while reducing integration time. In this paper we present the R2P framework detailing its architectural design and an example of robot platform developed with R2P. We also briefly discuss the flexibility of the approach and the easiness of reuse of its components.
Image quality assessment becomes essential for autonomous systems, where processing occurs on an acquired image and is then used for detection and recognition of objects. Images exhibiting low quality and captured in the presence of noise that are used as the basis for image recognition systems can dramatically impair the overall recognition system's performance. In this paper, we will present a new distance double variance color image quality measure that does not require a reference image in order to make its evaluation of the quality of an image. The Distance Doubling Variance measure differs from existing color image quality methods, which typically attempt to extend traditional grayscale image approaches for color images. Here, we utilize the color properties in the color space, where we evaluate the difference between two color pixels by computing the distance in the color space using different weights for each of the color components. Based on this distance, we calculate the double variance of the distance matrix. This matrix consists of the maximum distance of each pixel and its corresponding neighboring pixels. To demonstrate its performance, we use the TID-2013 database, which includes 24 different types of distortions for different kinds of images. The simulations are compared with state-of-the-art methods to show the new method has high agreement with human's visual system in many types of distortions.
Image corners encapsulate gradient changes in multiple directions. Therefore, corners are considered as efficient features for use in robotic navigation algorithms. Template based corner detection has a low computational complexity and is straightforward to implement. With the appropriate design of templates, satisfactory detection accuracy can also be achieved. In this paper, we introduce two new template based corner detection algorithms to be used to assist robot vision: the matching based corner detection, namely, MBCD; and the correlation based corner detection, namely, CBCD. These two approaches outperform existing template based approaches in the means that they reduce detection of spurious corners by considering ideal corners with at least two-pixel length on the corner arm directions. Experimental results show that the proposed algorithms detect essential corners for synthetic images and natural images satisfactorily according to human visual perception. We also examine the robustness of the two corner detection approaches in terms of the average repeatability and localization error. Since our approaches are computationally efficient, it makes these template based corner detection algorithms suitable for real time support in robotic applications. Comparisons with existing corner detection algorithms are also presented.
In our previous work, we introduced a new gait for humanoid robots called Ski-Type walking to improve stability performance for rough terrain walking. By the arms holding two canes to assist walking, the humanoid robot benefits from enlarged stability margin. With canes and feet touching the ground, a closed-chain system is formed where force/torque distribution among the canes and feet is not unique. We formulate and analyze both external and internal forces/torques in SkiType walking at its initial posture to determine the strategy for achieving optimal force/torque distribution. The result will enable our future study is designing dynamic gaits which specify not motion trajectories of the joints but also torques to according a desired criterion.
We describe the approach of Worcester Polytechnic Institute's (WPI) Robotics Engineering C Squad (WRECS) to the utility vehicle driving task at the Defense Advanced Research Projects Agency (DARPA) Robotics Challenge (DRC) Trials held in December 2013. WRECS was one of only seven teams to attempt the driving task, and the only team with an ATLAS robot to successfully drive the course. We implement a supervisory control system that allows the robot to control the speed of the vehicle, while the operator helps the robot steer the vehicle. Two different methods of estimating speed, using the LIDAR and stereo cameras, are presented, and the performance of the robot at the Trials is discussed.
This paper presents ANIM, a novel algorithm that uses angle-of-arrival (or bearing) measurements for relative positioning of networked, collaborating robots. The algorithm targets shortcomings of existing sensors (e.g., vulnerability of GPS to jamming) by providing a cheap, low-power alternative that can exploit existing, readily available communication equipment. The method is decentralized and iterative, with subgroups of three robots alternately estimating (i) orientation of the plane containing the robots and (ii) the direction of each edge between robots. Simulations demonstrate that ANIM converges reliably and provides accuracy sufficient for practical applications involving coordinated flying robots.
In order to access many spaces in human environments, mobile robots need to be adept at using doors: opening the door, traversing (i.e., passing through) the doorway, and possibly closing the door afterwards. The challenges in these problems vary with the type of door (push-/pull-doors, self-closing mechanisms, etc.) and type of door handle (knob, lever, crashbar, etc.) In addition, the capabilities and limitations of the robot can have a strong effect on the techniques and strategies needed for these tasks. We have developed a system that autonomously opens and traverses push- and pull-doors, with or without self-closing mechanisms, with knobs or levers, using an iRobot 510 PackBot® (a nonholonomic mobile base with a 5 degree-of-freedom arm) and a custom gripper with a passive 2 degree-of-freedom wrist. To the best of our knowledge, our system is the first to demonstrate autonomous door opening and traversal on the most challenging combination of a pull-door with a self-closing mechanism. In this paper, we describe the operation of our system and the results of our experimental testing.
This paper presents progress towards autonomous knot-tying in Robotic Assisted Minimally Invasive Surgery. While successful demonstrations of robotic knot-tying have been achieved, objective comparisons of competing approaches have been lacking. In this presentation we describe how to score a proposed procedure in terms of speed and volume. Applying the scoring metric has motivated an improved procedure for knot-tying, as well as a pathway to automated discovery for trajectory optimizations.
Soft actuators can be useful in human-occupied environments because of their adaptable compliance and light weight. We previously introduced a variation of fluidic soft actuators we call the reverse pneumatic artificial muscle (rPAM), and developed an analytical model to predict its performance both individually and while driving a 1 degree of freedom revolute joint antagonistically. Here, we expand upon this previous work, adding a correction term to improve model performance and using it to perform optimization on the kinematic module dimensions to maximize achievable joint angles. We also offer advances on the joint design to improve its ability to operate at these larger angles. The new joint had a workspace of around ±60°, which was predicted accurately by the improved model.
In this paper, we propose a novel method for obtaining product count directly from images recorded using a monocular camera mounted on a mobile robot. This has application in robot-based retail stock assessment problem where a mobile robot is used for monitoring the stock levels on the shelves of a retail store. The products are recognized by carrying out a nearest-neighbor search in the template feature space using a k-d tree. Unlike current approaches which only provide approximate stock level, we propose a method which can compute the exact number of discrete products visible in a given image. The product count is obtained by fitting bounding box around each product and removing them sequentially from the image. A second stage of grid-based search is carried out in the neighborhood of each detected product to detect new products which were missed out in the previous step. This detection is based on a confidence measure that includes various information such as histogram matching and spatial location. The efficacy of the proposed approach is demonstrated through experiments on different datasets obtained using robot camera as well as mobile phone camera. These results show that the robot-based retail stock assessment may become a viable alternative to the currently prevailing manual mode of carrying out these surveys.
An actively controlled drug delivery system (DDS) is an essential module to be included in the next generation of capsule endoscopy. Its development will allow physicians to perform non-invasive procedures and treat diseases in the digestive system. Despite many attempts to magnetically actuate internal permanent magnets (IPMs) embedded in prototype capsule robots to enhance their capabilities, further miniaturization and optimization of the IPMs are required to achieve more efficient torque transmission while minimizing the size of the IPMs. In this paper, we optimize the IPM's size to obtain a high magnetic torque that activates a DDS which is based on an overly miniaturized slider-crank mechanism. The IPM is optimized by means of analytical models. Our experimental results, which are in agreement with the analytical results, show that a high torque and force are generated on the piston of the DDS that expels drug out of a reservoir when an optimized IPM is embedded in the capsule robot.
To support studies in robot imitation learning, this paper presents a software platform, SMILE (Simulator for Maryland Imitation Learning Environment), specifically targeting tasks in which exact human motions are not critical. We hypothesize that in this class of tasks, object behaviors are far more important than human behaviors, and thus one can significantly reduce complexity by not processing human motions at all. As such, SMILE simulates a virtual environment where a human demonstrator can manipulate objects using GUI controls without body parts being visible to a robot in the same environment. Imitation learning is therefore based on the behaviors of manipulated objects only. A simple Matlab interface for programming a simulated robot is also provided in SMILE, along with an XML interface for initializing objects in the virtual environment. SMILE lowers the barriers for studying robot imitation learning by (1) simplifying learning by making the human demonstrator be a virtual presence and (2) eliminating the immediate need to purchase special equipment for motion capturing.
