Can supervise: YES
Woolfrey, J, Lu, W & Liu, D 2019, 'A Control Method for Joint Torque Minimization of Redundant Manipulators Handling Large External Forces', Journal of Intelligent and Robotic Systems.View/Download from: UTS OPUS or Publisher's site
© 2019, The Author(s). In this paper, a control method is developed for minimizing joint torque on a redundant manipulator where an external force acts on the end-effector. Using null space control, the redundant task is designed to minimize the torque required to oppose the external force, and reduce the dynamic torque. Furthermore, the joint motion can be weighted to factor in physical constraints such as joint limits, collision avoidance, etc. Conventional methods for joint torque minimization only consider the internal dynamics of the manipulator. If external forces acting on the end-effector are inadvertently implemented in to these control methods this could lead to joint configurations that amplify the resulting joint torque. The proposed control method is verified through two different case studies. The first case study involves simulation of high-pressure blasting. The second is a simulation of a manipulator lifting and moving a heavy object. The results show that the proposed control method reduces overall joint torque compared to conventional methods. Furthermore, the joint torque is minimized such that there is potential for a manipulator to execute certain tasks beyond its nominal payload capacity.
Wei, H, Lu, W, Zhu, P, Ferrari, S, Liu, M, Klein, RH, Omidshafiei, S & How, JP 2016, 'Information value in nonparametric Dirichlet-process Gaussian-process (DPGP) mixture models', AUTOMATICA, vol. 74, pp. 360-368.View/Download from: UTS OPUS or Publisher's site
Lu, W, Zhu, P & Ferrari, S 2016, 'A Hybrid-Adaptive Dynamic Programming Approach for the Model-Free Control of Nonlinear Switched Systems', IEEE TRANSACTIONS ON AUTOMATIC CONTROL, vol. 61, no. 10, pp. 3203-3208.View/Download from: UTS OPUS or Publisher's site
Lu, W, Zhang, G & Ferrari, S 2014, 'An Information Potential Approach to Integrated Sensor Path Planning and Control', IEEE TRANSACTIONS ON ROBOTICS, vol. 30, no. 4, pp. 919-934.View/Download from: UTS OPUS or Publisher's site
Lu, W, Zhang, G, Ferrari, S, Anderson, M & Fierro, R 2014, 'A particle-filter information potential method for tracking and monitoring maneuvering targets using a mobile sensor agent', Journal of Defense Modeling and Simulation, vol. 11, no. 1, pp. 47-58.View/Download from: Publisher's site
The problem of tracking and monitoring moving targets using mobile sensor agents (MSAs) is relevant to a variety of applications, including monitoring of endangered species, civilian security, and military surveillance. This paper presents a new information potential field approach for computing the motion plans and control inputs of a MSA, based on the feedback obtained from a modified particle filter used for tracking multiple moving targets in a region of interest. A modified particle filter is presented that implements a new sampling method based on supporting intervals of normal probability density functions. The method accounts for the latest sensor measurements by adapting a mixture representation of the target probability density functions (PDFs). The target motion is modeled as a semi-Markov jump process, such that the target PDFs, or the PDFs of the Markov parameters, can be updated based on real-time sensor measurements by a centralized processing unit or MSAs supervisor. A new information potential method is presented that computes an artificial potential function based on the output of the modified particle filter. Using this artificial potential, the sensors compute feedback control inputs that allow them to track and monitor a maneuvering target over time, using a bounded field of view (FOV). © 2012 The Society for Modeling.
Lu, W & Liu, D 2018, 'A Frequency-Limited Adaptive Controller for Underwater Vehicle-Manipulator Systems Under Large Wave Disturbances', 13th World Congress on Intelligent Control and Automation (WCICA), World Congress on Intelligent Control and Automation, IEEE, Changsha, China.View/Download from: UTS OPUS or Publisher's site
Wang, T, Lu, W & Liu, D 2018, 'Excessive Disturbance Rejection Control of Autonomous Underwater Vehicleusing Reinforcement Learning', 2018 Australasian Conference on Robotics and Automation, Lincoln, New Zealand.View/Download from: UTS OPUS
Small Autonomous Underwater Vehicles (AUV) in shallow water might not be stabilized well by feedback or model predictive control. This is because wave and current disturbances may frequently exceed AUV thrust capabilities and disturbance estimation and prediction models available are not sufficiently accurate. In contrast to classical model-free Reinforcement Learning (RL), this paper presents an improved RL for Excessive disturbance rejection Control (REC) that is able to learn and utilize disturbance behaviour, through formulating the disturbed AUV dynamics as a multi-order Markov chain. The unobserved disturbance behaviour is then encoded in the AUV state-action history of fixed length, its embeddings are learned within the policy optimization. The proposed REC is further enhanced by a base controller that is pre-trained on iterative Linear Quadratic Regulator (iLQR) solutions for a reduced AUV dynamic model, resulting in hybrid-REC. Numerical simulations on pose regulation tasks have demonstrated that REC significantly outperforms a canonical controller and classical RL, and that the hybrid-REC leads to more efficient and safer sampling and motion than REC.
Wang, T, Lu, W & Liu, D 2018, 'A Case Study: Modeling of A Passive Flexible Link on A Floating Platform for Intervention Tasks', 2018 13th World Congress on Intelligent Control and Automation, IEEE, Changsha, China.View/Download from: UTS OPUS
This paper focuses on modeling of a robotic system consisting of a floating platform and a passive flexible-link, which is subjected to three-dimensional large bending deformation during intervention tasks. It investigates the feasibility and efficacy of the quasi-Lagrangian approach and the Euler-Bernoulli beam assumption in modeling this system. Simulations and experiments were conducted to evaluate the model. Then the contact force was calculated with given external input force along with the pose and velocities of the robot, which is validated by the measurements obtained from force-torque sensors. It also found that the accelerations calculated from the model have some deviation from the results obtained from a tracking system.
Lu, W & Liu, D 2017, 'Active Task Design in Adaptive Control of Redundant Robotic Systems', Australasian Conference on Robotics and Automation 2017, Australasian Conference on Robotics and Automation, ARAA, Sydney Australia, pp. 1-6.View/Download from: UTS OPUS
This paper seeks for possibilities of using robots' kinematic redundancy to excite the system persistently, through actively designing a secondary task in the null space of a primary task. Resulted parameter convergence in adaptive control leads to better system stability and performance.
A measure in Grassmannian, referred to as Subspace Discrepancy Measure (SDM), is proposed for evaluating the additional benefit from the secondary task in converging unknown parameters to their true values. This measure evaluates the angles among subspaces that the parameter estimations are converging to, given different secondary tasks. The subspaces are obtained from Principal Component Analysis (PCA) on a small amount of samples of parameter estimations. The SDM is used to determine the choice of the secondary task online through a trial-and-evaluation procedure actively. Numerical simulations demonstrated that the secondary task chosen by SDM enhances the parameter convergence.
© 2015 IEEE. This paper presents a multi-layer reproducing kernel Hilbert space (RKHS) approach for probability distribution to real and probability distribution to function regressions. The approach maps the distributions into RKHS by distribution embeddings and, then, constructs a multi-layer RKHS within which the multi-kernel distribution regression can be implemented using an existing kernel regression algorithm, such as kernel recursive least squares (KRLS). The numerical simulations on synthetic data obtained via Gaussian mixtures show that the proposed approach outperforms existing probability distribution (DR) regression algorithms by achieving smaller mean squared errors (MSEs) and requiring less training samples.
Wei, H, Lu, W, Zhu, P, Ferrari, S, Klein, RH, Omidshafiei, S & How, JP 2014, 'Camera control for learning nonlinear target dynamics via Bayesian nonparametric Dirichlet-process Gaussian-process (DP-GP) models', IEEE International Conference on Intelligent Robots and Systems, pp. 95-102.View/Download from: Publisher's site
© 2014 IEEE. This paper presents a camera control approach for learning unknown nonlinear target dynamics by approximating information value functions using particles that represent targets' position distributions. The target dynamics are described by a non-parametric mixture model that can learn a potentially infinite number of motion patterns. Assuming that each motion pattern can be represented as a velocity field, the target behaviors can be described by a non-parametric Dirichlet process-Gaussian process (DP-GP) mixture model. The DP-GP model has been successfully applied for clustering time-invariant spatial phenomena due to its flexibility to adapt to data complexity without overfitting. A new DP-GP information value function is presented that can be used by the sensor to explore and improve the DP-GP mixture model. The optimal camera control is computed to maximize this information value function online via a computationally efficient particle-based search method. The proposed approach is demonstrated through numerical simulations and hardware experiments in the RAVEN testbed at MIT.
Wei, H, Lu, W, Zhu, P, Huang, G, Leonard, J & Ferrari, S 2014, 'Optimized visibility motion planning for target tracking and localization', IEEE International Conference on Intelligent Robots and Systems, pp. 76-82.View/Download from: Publisher's site
© 2014 IEEE. This paper presents a visibility-based method for planning the motion of a mobile robotic sensor with bounded field-of-view to optimally track a moving target while localizing itself. The target and robot states are estimated from online sensor measurements and a set of a priori known landmarks, using an extended Kalman filter (EKF), and thus the proposed method is applicable to robots without a global positioning system. It is shown that the problem of optimizing the target tracking and robot localization performance is equivalent to optimizing the visibility or probability of detection in the EKF framework under mild assumptions. The control law that maximizes the probability of detection for a robotic sensor with a sector-shaped field-of-view (FoV) is derived as a function of the robot heading and aperture. Simulations have been conducted on synthetic experiments and the results show that the optimized-visibility approach is effective at avoiding target loss, and outperforms a state-of-the-art potential method based on robot trailer models .
Bellini, AC, Lu, W, Naldi, R & Ferrari, S 2014, 'Information driven path planning and control for collaborative aerial robotic sensors using artificial potential functions', Proceedings of the American Control Conference, pp. 590-597.View/Download from: Publisher's site
A path planning and control method based on adaptive potential functions is presented for a group of unmanned aerial vehicles (UAVs) equipped with onboard sensors, and deployed to search and classify multiple targets. The proposed method plans the motion of the UAVs to support a primary sensing objective that, in this case, is to maximize the classification performance of the sensor measurements gathered by the UAVs over time. An adaptive potential function approach originally developed for ground robots is modified and employed as a guidance law for a class of rotary-wing UAVs that must also avoid obstacles located in a three-dimensional workspace. The simulation results show that, by this approach, a single UAV is capable of visiting targets that offer the best tradeoff between distance and measurement information value. Furthermore, simulations involving multiple UAVs deployed to classify the same set of targets show that, by this approach, there emerge a cooperative behavior by which the UAVs can react, as a group, to the targets' classification uncertainties. © 2014 American Automatic Control Council.
Lu, W & Ferrari, S 2013, 'An approximate dynamic programming approach for model-free control of switched systems', Proceedings of the IEEE Conference on Decision and Control, pp. 3837-3844.View/Download from: Publisher's site
Several approximate dynamic programming (ADP) algorithms have been developed and demonstrated for the model-free control of continuous and discrete dynamical systems. However, their applicability to hybrid systems that involve both discrete and continuous state and control variables has yet to be demonstrated in the literature. This paper presents an ADP approach for hy- brid systems (hybrid-ADP) that obtains the optimal control law and discrete action sequence via online learning. New recursive relationships for hybrid-ADP are presented for switched hybrid systems that are possibly nonlinear. In order to demonstrate the ability of the proposed ADP algorithm to converge to the optimal solution, the approach is demonstrated on a switched, linear hybrid system with a quadratic cost function, for which there exists an analytical solution. The results show that the ADP algorithm is capable of converging to the optimal switched control law, by minimizing the cost-to-go online, based on an observable state vector. ©2013 IEEE.
Zielinski, DJ, Kopper, R, McMahan, RP, Lu, W & Ferrari, S 2013, 'Intercept tags: Enhancing intercept-based systems', Proceedings of the ACM Symposium on Virtual Reality Software and Technology, VRST, pp. 263-266.View/Download from: Publisher's site
In some virtual reality (VR) systems, OpenGL intercept methods are used to capture and render a desktop application's OpenGL calls within an immersive display. These systems often suffer from lower frame rates due to network bandwidth limitations, implementation of the intercept routine, and in some cases, the intercepted application's frame rate. To mitigate these issues and to enhance intercept-based systems in other ways, we present intercept tags, which are OpenGL geometries that are interpreted instead of rendered. We have identified and developed several uses for intercept tags, including hand-off interactions, display techniques, and visual enhancements. To demonstrate the value of intercept tags, we conducted a user study to compare a simple virtual hand technique implemented with and without intercept tags. Our results show that intercept tags significantly improve user performance and experience.
Zielinski, DJ, McMahan, RP, Lu, W & Ferrari, S 2013, 'ML2VR: Providing MATLAB users an easy transition to virtual reality and immersive interactivity', Proceedings - IEEE Virtual Reality, pp. 83-84.View/Download from: Publisher's site
MATLAB is a popular computational system and programming environment that is used in numerous engineering and science programs in the United States. One feature of MATLAB is the capability to generate 3D visualizations, which can be used to visualize scientific data or even to simulate engineering models and processes. Unfortunately, MATLAB provides only limited interactivity for these visualizations. As a solution to this problem, we have developed a software system that easily integrates with MATLAB scripts to provide the capability to view visualizations and interact with them in virtual reality (VR) systems. We call this system 'ML2VR' and expect it will introduce more users to VR by enabling a large population of MATLAB programmers to easily transition to immersive systems. We will describe the system architecture of ML2VR and report on a successful case study involving the use of ML2VR. © 2013 IEEE.
Lu, W, Ferrari, S, Fierro, R & Wettergren, TA 2012, 'Approximate dynamic programming recurrence relations for a hybrid optimal control problem', Proceedings of SPIE - The International Society for Optical Engineering.View/Download from: Publisher's site
This paper presents a hybrid approximate dynamic programming (ADP) method for a hybrid dynamic system (HDS) optimal control problem, that occurs in many complex unmanned systems which are implemented via a hybrid architecture, regarding robot modes or the complex environment. The HDS considered in this paper is characterized by a well-known three-layer hybrid framework, which includes a discrete event controller layer, a discrete-continuous interface layer, and a continuous state layer. The hybrid optimal control problem (HOCP) is to nd the optimal discrete event decisions and the optimal continuous controls subject to a deterministic minimization of a scalar function regarding the system state and control over time. Due to the uncertainty of environment and complexity of the HOCP, the cost-to-go cannot be evaluated before the HDS explores the entire system state space; as a result, the optimal control, neither continuous nor discrete, is not available ahead of time. Therefore, ADP is adopted to learn the optimal control while the HDS is exploring the environment, because of the online advantage of ADP method. Furthermore, ADP can break the curses of dimensionality which other optimizing methods, such as dynamic programming (DP) and Markov decision process (MDP), are facing due to the high dimensions of HOCP. © 2012 Copyright Society of Photo-Optical Instrumentation Engineers (SPIE).
Lu, W, Zhang, G & Ferrari, S 2012, 'A comparison of information theoretic functions for tracking maneuvering targets', 2012 IEEE Statistical Signal Processing Workshop, SSP 2012, pp. 149-152.View/Download from: Publisher's site
Several information theoretic functions have been proposed in the literature to assess the information value of sensor measurements a posteriori, that is, after measurements have been obtained from one or more targets. Sensor planning algorithms, however, require that the value of future sensor measurements be computed a priori, based on available models and prior information. An approach was recently presented by the authors for estimating the expected information value of future sensor measurements in target classification problems. The approach derives expected information theoretic functions from probabilistic models of the sensors and the targets, conditioned on prior information. In this paper, the approach is extended to the problem of sensor planning for tracking maneuvering targets. The approach is illustrated for a sensor that obeys an exponential power law model of received isotropic energy, and a target that obeys a Markov motion model. The performance of five information theoretic functions is compared through numerical simulations, and the results show that the objective function based on conditional mutual information leads to the most effective sensor planning strategy. © 2012 IEEE.
Ferrari, S, Anderson, M, Fierro, R & Lu, W 2011, 'Cooperative navigation for heterogeneous autonomous vehicles via approximate dynamic programming', Proceedings of the IEEE Conference on Decision and Control, pp. 121-127.View/Download from: Publisher's site
Unmanned ground and aerial vehicles are becoming crucial to many applications because of their ability to assist humans in carrying out dangerous missions. These vehicles can be viewed as networks of heterogeneous unmanned robotic sensors with the goal of exploring complex environments, to search for and, possibly, pursue moving targets. The robotic vehicle performance can be greatly enhanced by implementing future sensor actions intelligently, based both on prior knowledge and on the information obtained by the sensors on line. In this paper, we present an approximate dynamic programming (ADP) approach to cooperative navigation for heterogeneous sensor networks. The mobile sensor network consists of a set of robotic sensors modeled as hybrid systems with processing capabilities. The goal of the ADP algorithm is to coordinate a team of heterogeneous autonomous vehicles (i.e., ground robot and quadrotor UAV) to navigate within an obstacle populated environment while satisfying collision avoidance constraints and searching for stationary and mobile targets. It is assumed that the ground vehicle has a small sensor footprint with high resolution. The quadrotor, on the other hand, has a large sensor field-of-view but low resolution. The UAV provides a low resolution look-ahead map to the ground robot which in turn uses this information to plan its actions. The proposed navigation strategy combines artificial potential functions for target pursuing with ADP for learning C-obstacles on line. The efficacy of the proposed methodology is verified through numerical simulations. © 2011 IEEE.
Lu, W, Zhang, G, Ferrari, S, Fierro, R & Palunko, I 2011, 'An information potential approach for tracking and surveilling multiple moving targets using mobile sensor agents', Proceedings of SPIE - The International Society for Optical Engineering.View/Download from: Publisher's site
The problem of surveilling moving targets using mobile sensor agents (MSAs) is applicable to a variety of fields, including environmental monitoring, security, and manufacturing. Several authors have shown that the performance of a mobile sensor can be greatly improved by planning its motion and control strategies based on its sensing objectives. This paper presents an information potential approach for computing the MSAs' motion plans and control inputs based on the feedback from a modified particle filter used for tracking moving targets. The modified particle filter, as presented in this paper implements a new sampling method (based on supporting intervals of density functions), which accounts for the latest sensor measurements and adapts, accordingly, a mixture representation of the probability density functions (PDFs) for the target motion. It is assumed that the target motion can be modeled as a semi-Markov jump process, and that the PDFs of the Markov parameters can be updated based on real-time sensor measurements by a centralized processing unit or MSAs supervisor. Subsequently, the MSAs supervisor computes an information potential function that is communicated to the sensors, and used to determine their individual feedback control inputs, such that sensors with bounded field-of-view (FOV) can follow and surveil the target over time. © 2011 Copyright Society of Photo-Optical Instrumentation Engineers (SPIE).
Lu, W, Zhang, G & Ferrari, S 2010, 'A randomized hybrid system approach to coordinated robotic sensor planning', Proceedings of the IEEE Conference on Decision and Control, pp. 3857-3864.View/Download from: Publisher's site
This paper proposed a randomized hybrid system approach for planning the paths and measurements of a network of robotic sensors deployed for searching and classifying objects in a partially-observed environment containing multiple obstacles and multiple targets. The sensor planning problem considered in this paper consists of coordinating and planning the motions of each robot, equipped with two on-board sensing capabilities. One sensing capability is assumed to have low classification accuracy and a large field of view (FOV), and the other is assumed to have high classification accuracy and a smaller FOV. A sampling function for rapidly-exploring trees is presented that takes into account both sensor measurements of obstacle locations and robot configuration and velocity to generate new milestones for the tree online. The tree expansion also takes into account the expected information value of the targets, represented by conditional mutual information, in order to favor expansions toward targets with higher measurement benefit. The proposed method is implemented and demonstrated on a network of robotic sensors simulated using the 3D physics-based robotics software packages Player/Stage/Gazebo. ©2010 IEEE.