Detection of interest features is a fundamental pre-processing task for visual odometry and visual simultaneous localization and mapping. The combination of line segments and corners is a trend for those applications, and it is challenging to meet real-time constraints. To pursue this goal, this letter presents a novel points and line points feature detection (PLPD) algorithm designed to be highly efficient and accurate for real-time VO-based systems. The proposed algorithm adopts a novel Multi-level Edge Detector (MED) based on gradient intensity and structure tensor that is not confined solely to extracting corners, but rather, it exploits feature-line points properties to extract long-line segments directly without merging collinear fragment segments. The main focus of this work is to fulfill the real-time and embeddability constraints of our proposed method by adopting the GPU-aware software design for a suitable implementation in GPU-based embedded heterogeneous architecture. Based on this, we exhibit extensive benchmarking with other state-of-the-art algorithms for feature-line segment extraction and corner detection, showing the efficiency of the proposed algorithm and ensuring a real-time performance up to 33 FPS using our dataset.

In this paper we report the safety-oriented framework for controlling the torque in the case of robots with high reduction gears and having no joint torque feedback. This kind of robots suffer from high joint friction and low backdrivability, requiring high gains and integral feedback, which can be dangerous. Our optimization-based framework includes feasibility and safety features borrowed from position control, and we introduce novel ones. We show how we limit the integral terms using a QP-based anti-windup which produces the optimal torque that maintains the best performances under safety limits. We show also a new controller for null-space compliance, providing strong guarantees of convergence in the task-space and ignoring the corresponding null-space where the robot can be moved freely. We validate these features with experiments on one 9 DoF arm of the robot HRP-5P performing a Cartesian task, and then a dual Cartesian / admittance task.

We propose a control framework for torque controlled humanoid robots that efficiently minimizes the tracking error in a Quadratic Programming (QP)formulated as multiobjective weighted tasks with constraints. It results in an optimal dynamically-feasible reference that can be tracked robustly, with exponential convergence, without joint torque feedback, in the presence of non modelled torque bias and low-frequency bounded disturbances. This is achieved by introducing integral gains in a Lyapunov-stable torque control, which exploit the passivity properties of the dynamical model of the robot and their effect on the dynamic constraints of the QP solver. The robustness of this framework is demonstrated in simulation by commanding our robot, the HRP-5P, to achieve simultaneously several objectives in the configuration and the Cartesian spaces, in the presence of non-modeled static and kinetic joint friction, as well as an uncertain torque scale.

This paper proposes simple and efficient algorithms for implementation of attitude estimation and control based on data fusion using complementary filters taking into account sensors dynamics. First of all, we propose a passive form of the filter by fusing the measured inertial vectors and the gyro measurements in order to reconstruct real inertial vectors which can be used with any algebraic algorithm (TRIAD, QUEST, etc.) that leads to globally asymptotic attitude estimation. Thereafter, the same principle of data fusion is used to address the problem of attitude stabilization. Then, instead of using direct raw measurements in control law we propose a new solution that leads to accurate estimation of inertial vectors by using complementary filters based on sensors dynamics. The stability analysis of the error dynamics based on Lyapunov method proved that almost all trajectories converge asymptotically to the desired equilibrium point. Simulation results show the effectiveness and the performance of the proposed solutions.

In this paper a new method for attitude estimation of rigid body using line-based vision and a Multiplicative Extended Kaiman Filter (MEKF) is developed. A vision-based line-tracking algorithm that allows to detect and to track points and lines along sequence of images without drift is used. From this algorithm, we can get an implicit measure of the lines direction. The latter are then fused with gyro measurements using an observer designed on SO (3) in order to estimate attitude with gyro bias compensation. The gain matrices of the proposed observer are determined based on continuous-time MEKF. The problem of sign ambiguity related to the implicit measure of direction lines is addressed and a correction factor is used to remove this ambiguity. Simulation results has been presented to show the effectiveness of the proposed approach.

We propose a velocity-free attitude stabilization scheme in which neither the angular velocity nor the instantaneous measurements of the attitude are used in the feedback, only body vector measurements are needed. To overcome the lack of angular velocity, a first order linear auxiliary system based directly on these vector measurements is introduced. Almost global asymptotic stability results are obtained. Also, in order to adjust properly the gains of the controller, an analysis of their effect on the closed-loop dynamics was performed. The effectiveness and performance of the proposed solution are illustrated via simulation results where some comparison with existing previous work are given.

This paper addresses the problem of estimation of attitude of accelerated rigid body vehicles moving in 3D-space using IMU and data related to measurement of velocity by GPS. We propose a new observer for attitude estimation based on proportional-integral (PI) compensator along with discontinuous switching function in the estimation of velocity to compensate the effect of linear acceleration. The convergence analysis of the proposed observer is studied using Barbalat's lemma based on a Lyapunov like positive definite function to guarantee asymptotic convergence. The performance of the observer is illustrated by simulation results which show the convergence behaviour of the proposed observer.

In this paper, an original method about heavy vehicles rollover risk prediction is presented and validated experimentally. It is based on the calculation of the LTR (load transfer ratio) which depends on the estimated vertical forces using high order sliding mode observers. The validation tests were carried out on an instrumented truck rolling on the road at various speeds and lane-change manoeuvres. Many scenarios have been experienced: driving on straight line, curve line and zigzag to emphasize the rollover phenomenon and its prediction to set off an alarm to the driver.

This paper presents a nonlinear observer algorithm for attitude estimation that improves the quality of measures obtained by using low-cost inertial measurements (IMU). It is based on sliding mode observer that provides both the estimates of the gyro-bias and the actual attitude of the rigid body. The algorithm was developed in order to address the well-known problem of the weak dynamics of the tilt sensors and magnetometers, which can be modeled by low pass filters, and of the measurement bias of the gyros. In its design the observer uses the real measurements given by the low-cost attitude sensors (inclinometers and magnetometers) and the gyros, the filters modeling the sensors and the kinematics equation of the rigid body. The asymptotic convergence of the estimation of the attitude and bias-gyros was proven using Lyapunov stability method. The effectiveness of the algorithm has been shown from experimental tests using a rotary platform equipped with several sensors with axes of rotation coincide with orientation of the rigid body. Also, tests for comparison with a linear complementary filter are given.

This paper presents the design of a backstepping controller using sliding mode estimation technique which aims to simplify the control procedure. This approach, applied to a quadrotor unmanned aerial vehicle, differs from standard backstepping in that the virtual control inputs are designed based on estimates of the previous virtual control inputs. This eliminates the need to take derivatives of the system dynamics, which simplifies the control law. The estimation design is based on the exact second-order sliding mode differentiator. The controller objective is to achieve good tracking of desired positions and yaw angle while keeping the stability of the pitch and roll angles. The design methodology is based on the Lyapunov stability. Simulation results demonstrate the effectiveness of the proposed approach.

This paper presents a nonlinear dynamic model for a quadrotor helicopter in a form suited for backstepping control design. Due to the under-actuated property of quadrotor helicopter, the controller can set the helicopter track three Cartesian positions (x,y,z) and the yaw angle to their desired values and stabilize the pitch and roll angles. The system has been presented into three interconnected subsystems. The first one representing the under-actuated subsystem, gives the dynamic relation of the horizontal positions (x,y) with the pitch and roll angles. The second fully-actuated subsystem gives the dynamics of the vertical position z and the yaw angle. The last subsystem gives the dynamics of the propeller forces. A backstepping control is presented to stabilize the whole system. The design methodology is based on the Lyapunov stability theory. Various simulations of the model show that the control law stabilizes a quadrotor with good tracking

In this paper, a mixed robust feedback linearization with linear GH controller is applied to a non linear quadrotor unmanned aerial vehicle. An actuator saturation and constrain on state space output are introduced to analyse the worst case of control law design. The results show that the overall system becomes robust when weighting functions are chosen judiciously. Performance issues of the controller are illustrated in a simulation study that takes into account parameter uncertainties and external disturbances as well as measurement noise.

A nonlinear dynamic model for a quadrotor unmanned aerial vehicle is presented with a new vision of state parameter control which is based on Euler angles and open loop positions state observer. This method emphasizes on the control of roll, pitch and yaw angle rather than the translational motions of the UAV. For this reason the system has been presented into two cascade partial parts, the first one relates the rotational motion whose the control law is applied in a closed loop form and the other one reflects the translational motion. A dynamic feedback controller is developed to transform the closed loop part of the system into linear, controllable and decoupled subsystem. The wind parameters estimation of the quadrotor is used to avoid more sensors. Hence an estimator of resulting aerodynamic moments via Lyapunov function is developed. Performance and robustness of the proposed controller are tested in simulation.

The paper deals with a neural network adaptive controller designed for constrained redundant robot manipulators. The controller has been determined using extended cartesian space to ensure minimum joint positions of the robot and to take into account mechanical constraints like joint limitations. The proposed approach guarantees a good minimization of any performance criterion, subject to either equality or inequality constraints while achieving the end-effector task. It verifies the repeatability property for closed or cyclic trajectories and avoids the computation of the inverse or pseudoinverse Jacobian extended matrix. Several neural networks are used to approximate separately the elements of the dynamical model of the robot manipulator written in cartesian space. Adaptation laws are derived for each network to ensure stability of the closed loop system. Simulations results demonstrate a good performance of the proposed controller.

We propose a new solution to the inverse kinematic problem of redundant robots subject to a set of criteria and constraints. First, a study of existing methods leads us to develop an on-line algorithm based on an adaptive neural network. This solution needs only a few iterations to converge, offers substantially better accuracy, verifies the repeatability propriety for the closed trajectory and avoids the computation of the inverse or pseudoinverse Jacobian matrix. Our approach guarantees a good minimization of any performance criterion subject to either equality or inequality constraints while achieving the end-effector task. Then, our method can solve the inverse kinematic problem of a redundant robot for it to follow a desired trajectory while avoiding moved or fixed obstacles.

A neural network adaptive force controller is proposed for a real hydraulic system. The dynamic model of this system is highly non-linear and very complex to obtain. Thus, it is considered as a black box, and a priori identification becomes necessary. A neural network is used to approximate the model, then a controller using the Lyapunov approach is designed. The neural network parameters are updated online according to an adaptation algorithm obtained via stability analysis. The performance of the proposed neural network controller is validated on an experimental plant.

A neural network-based robust adaptive tracking controller is proposed for a class of nonlinear multi-input multi-output (MIMO) systems. The nonlinear system is treated as a partially known system. The known dynamics are used to design a nominal feedback controller, and a neural network-based adaptive compensator is designed to compensate the effects of the system uncertainties. By this scheme, both strong robustness with respect to unknown dynamics and asymptotic convergence to zero of the output tracking error are obtained.