Although sample-based model predictive control (MPC), such as model predictive path integral control (MPPI), are well suited to manage complex tasks like tracking a teleoperated robot while maintaining constraints and avoiding obstacles, it can be challenging to design the MPPI input sampling to achieve precision tracking. The main contribution of this work is to enable precision tracking with MPPI-type sampling methods by (i) sampling the system's reference outputs and (ii) using inversion based control to correct for system dynamics. An advantage of the proposed reference-output-sampled MPPI (oMPPI) is that the selected reference output's sample distribution can reflect the desired output of the system, such as the trajectory of the teleoperated robot in the active vision application. The proposed oMPPI is applied to a crane-robot active vision system for confined space inspection during aircraft wing manufacturing, enabling a teleoperated manipulator to navigate around in-wing structures. Teleoperation experiments show that oMPPI sampling increases tracking precision of a teleoperated manipulator by 22% and reduces camera oscillations by 65% when compared to MPPI sampling without inversion.

Leader-follower -type platooning can aid human management of multiple robots through complex environments, where the human manages constraints and obstacle-avoidance of the lead robot and the other robots follow the same human-generated leader trajectory. However, the spacing error between robots in the platoon should be small to ensure that each robot in the platoon maintains visual contact with its predecessor during tight maneuvers around obstacles. The main challenge is that current decentralized constant-spacing platoons require robot-to-robot communication, to avoid string instability that would increase spacing error along the length of the platoon. Such communication is not always available when ad-hoc platoons are formed without communication infrastructure, or when such broadcast communication is not permissible due to security or jamming concerns. The primary contribution of this paper is the development of a novel string-stable platooning strategy, that enables arbitrarily small spacing error, provided the sensing rate is sufficiently high, using a Delayed Self-Reinforcement (DSR) method. This is the first algorithm that shows arbitrarily small spacing error for constant-spacing platoons without communication. Additionally, simulation results show that the DSR approach results in a 85% reduction in the longitudinal target spacing error for autonomous platoons when compared to the baseline without DSR, which aids in reducing task completion time by 44% in human-led ($N=10$) platoons.

Ergodic exploration has spawned a lot of interest in mobile robotics due to its ability to design time trajectories that match desired spatial coverage statistics. However, current ergodic approaches are for continuous spaces, which require detailed sensory information at each point and can lead to fractal-like trajectories that cannot be tracked easily. This paper presents a new ergodic approach for graph-based discretization of continuous spaces. It also introduces a new time-discounted ergodicity metric, wherein early visitations of information-rich nodes are weighted more than late visitations. A Markov chain synthesized using a convex program is shown to converge more rapidly to just ergodicity than the traditional fastest mixing Markov chain. The resultant ergodic traversal method is used within a hierarchical framework for active inspection of confined spaces with the goal of detecting anomalies robustly using SLAM-driven Bayesian hypothesis testing. Experiments on a ground robot show the advantages of this framework over typical continuous space ergodic planners as well as greedy and random exploration methods for left-behind foreign object debris detection in a ballast tank.

Visual inspection of confined spaces such as aircraft wings is ergonomically challenging for human mechanics. This work presents a novel crane robot that can travel the entire span of the aircraft wing, enabling mechanics to perform inspection from outside of the confined space. However, teleoperation of the crane robot can still be a challenge due to the need to avoid obstacles in the workspace and potential oscillations of the camera payload. The main contribution of this work is to exploit the differential flatness of the crane-robot dynamics for designing reduced-oscillation, collision-free time trajectories of the camera payload for use in teleoperation. Autonomous experiments verify the efficacy of removing undesired oscillations by 89%. Furthermore, teleoperation experiments demonstrate that the controller eliminated collisions (from 33% to 0%) when 12 participants performed an inspection task with the use of proposed trajectory selection when compared to the case without it. Moreover, even discounting the failures due to collisions, the proposed approach improved task efficiency by 18.7% when compared to the case without it.

One of the goals of active information acquisition using multi-robot teams is to keep the relative uncertainty in each region at the same level to maintain identical acquisition quality (e.g., consistent target detection) in all the regions. To achieve this goal, ergodic coverage can be used to assign the number of samples according to the quality of observation, i.e., sampling noise levels. However, the noise levels are unknown to the robots. Although this noise can be estimated from samples, the estimates are unreliable at first and can generate fluctuating values. The main contribution of this paper is to use simulated annealing to generate the target sampling distribution, starting from uniform and gradually shifting to an estimated optimal distribution, by varying the coldness parameter of a Boltzmann distribution with the estimated sampling entropy as energy. Simulation results show a substantial improvement of both transient and asymptotic entropy compared to both uniform and direct-ergodic searches. Finally, a demonstration is performed with a TurtleBot swarm system to validate the physical applicability of the algorithm.

Trajectory-sampling-based techniques, such as model predictive path integral control (MPPI), are well suited to solve complex optimization problems with numerous weighted cost terms and non-smooth constraints. However, it can be challenging to design the input-sampling distribution to fully capture and correct the effects of system dynamics to enable precision output tracking. This challenge can be resolved with the output-sampled MPPI (oMPPI) approach, which first augments the system with its inverse to generate inputs that account for the system dynamics and achieve accurate output tracking, and second applies the standard MPPI approach to the augmented system with the inverse. The main contribution of this work is to show that optimality of MPPI is preserved with the oMPPI formulation when sampling the system’s reference outputs — rather than sampling the inputs as in standard MPPI. This theoretical development validates experimental improvements shown in prior work on oMPPI.

This work considers human-robot collaborative transportation (co-transport) of flexible objects using force/torque measurement at the robot end-effector (which is standard in industrial robots) without using external sensors such as cameras. The challenge is that standard admittance control, typically used for rigid-object co-transport, can lead to low gain margins and deformation of the object - large deformations can damage flexible objects. The main contribution of this work is to increase the gain margin of admittance controllers by compensating for the flexible-object dynamics and thereby reduce the structural deformation during cotransport. Experimental results with the proposed approach show that the deformation can be reduced by approximately half (45%) when compared to standard admittance control.

Payload swing during rapid slewing of mobile cranes poses a safety risk, as it generates overturning moments that can lead to tip-over accidents of mobile cranes. Currently, to limit the risk of tip-over, mobile crane operators are forced to either reduce the slewing speed (which lowers productivity) or reduce the load being carried to reduce the induced moments. Both of these approaches reduce productivity. This paper seeks to enable rapid slewing without compromising safety by applying input shaping to the crane-slewing commands generated by the operator. A key advantage of this approach is that the input shaper requires only the information about the rope length, and does not require detailed mobile crane dynamics. Simulations and experiments show that the proposed method reduces residual payload swing and enables significantly higher slewing speeds without tip over, reducing slewing completion time by at least 38

Ergodic exploration has spawned a lot of interest in mobile robotics due to its ability to design time trajectories that match desired spatial coverage statistics. However, current ergodic approaches are for continuous spaces, which require detailed sensory information at each point and can lead to fractal-like trajectories that cannot be tracked easily. This paper presents a new ergodic approach for graph-based discretization of continuous spaces. It also introduces a new time-discounted ergodicity metric, wherein early visitations of information-rich nodes are weighted more than late visitations. A Markov chain synthesized using a convex program is shown to converge more rapidly to time-discounted ergodicity than the traditional fastest mixing Markov chain. The resultant ergodic traversal method is used within a hierarchical framework for active inspection of confined spaces with the goal of detecting anomalies robustly using SLAM-driven Bayesian hypothesis testing. Experiments on a ground robot show the advantages of this framework over three continuous space ergodic planners as well as greedy and random exploration methods for left-behind foreign object debris detection in a ballast tank.

During manufacturing processes, such as clamping and drilling of elastic structures, it is essential to maintain tool-workpiece normality to minimize shear forces and torques, thereby preventing damage to the tool or the workpiece. The challenge arises in making precise model-based predictions of the relatively large deformations that occur as the applied normal force (e.g., clamping force) is increased. However, precision deformation predictions are essential for selecting the optimal robot pose that maintains force normality. Therefore, recent works have employed force-displacement measurements at each work location to determine the robot pose for maintaining tool normality. Nevertheless, this approach, which relies on local measurements at each work location and at each gradual increment of the applied normal force, can be slow and consequently time prohibitive. The main contributions of this work are: (i) to use Gaussian process (GP) methods to learn the robot-pose map for force normality at unmeasured workpiece locations; and (ii) to use active learning to optimally select and minimize the number of measurement locations needed for accurate learning of the robot-pose map. Experimental results show that the number of data points needed with active learning is 77.8% less than the case with a benchmark linear positioning learning for the same level of model precision. Additionally, the learned robot-pose map enables a rapid increase of the normal force at unmeasured locations on the workpiece, reaching force-increment rates up to eight times faster than the original force-increment rate when the robot is learning the correct pose.

This work investigates the tracking of dynamic trajectories using consensus-based decentralized algorithms. Dynamic average consensus (DAC) algorithms seek to improve tracking of the average signal by increasing the convergence rate, e.g., using accelerated versions of standard DAC algorithms. The main contribution of this work is to show that improving the convergence rate doesn't necessarily ensure improved tracking of the average signal, especially during rapid changes in the measured signals. Additionally, this work also presents a DAC-based delayed self reinforcement (DSR) approach to improve the tracking of the average signal rather than the convergence rate. The proposed DSR-DAC algorithm, which approximates the ideal centralized DAC, improves tracking of the average signal by only using already-available (current and past), local information from the network and sensed reference signals. Numerical simulation results for a multi-mobile agent tracking problem demonstrate substantial improvement with 91% and 84% reduction in tracking error using the proposed DSR-DAC approach when compared to the standard and accelerated DAC algorithms, respectively.

The advent of easy access to large amount of data has sparked interest in directly developing the relationships between input and output of dynamic systems. A challenge is that in addition to the applied input and the measured output, the dynamics can also depend on hidden states that are not directly measured. In general, it is unclear what type of data, such as past input and or past output is needed, to learn inverse operators (that predict the input needed to track a desired output for control purposes) with a desired precision. The main contribution of this work is to show that, irrespective of the selected model, removing the hidden-state dependence and achieving a desired precision of inverse operators require (i) a sufficiently-long past history of the output and (ii) sufficiently-precise estimates of the output's instantaneous time derivatives that are necessary and sufficient for linear systems, and under some conditions, for nonlinear systems. This insight, about the required observables (output history and derivative) for removing the hidden-state dependence and achieving precision, is used to develop a data-enabled algorithm to learn the inverse operator for multi-input multi-output square systems. Simulation examples are used to illustrate that neural nets (with universal approximation property) can learn the inverse operator with sufficient precision only if the required observables, identified in this work, are included in training.

Effective decision-making by robotic networks to collectively achieve coupled objectives requires accurate information communication. Typically, all environmental information (e.g., task values) are assumed to be centrally known by decision-making algorithms used for solving problems such as task assignment (TA). A challenge is that the task values might only emerge after information about the environment (that might only be available to some agents) is accurately shared by the agents. However, existing decentralized communication methods, such as the standard (consensus) method can lead to distortion in the information as it diffuses between distant agents, resulting in large settling times for task values, which in turn can lead to ineffective decisions—even if consensus is eventually achieved in the TA. The main contribution of this work is to improve decision-making in robot networks by improving the accuracy of shared environmental information needed to compute task values using a noise-suppressing, delayed-self-reinforcement (DSR) approach that reduces the transient information distortion. DSR approximates the ideal, distortion-free, centralized information sharing using only decentralized information sharing, and does not require changes to the network topology or increased communication bandwidth. Furthermore, this work develops communication-error bounds for DSR in terms of the second time derivative of the communicated information. Experimental results show substantial improvement in the accuracy of the information with a 95% and 88% error reduction in position and speed information with the proposed method, respectively, when compared to the standard method, resulting in an 88% improvement in settling times for task values and 100% successful task capture rate with the proposed information communication method as opposed to loss of task capture using the standard method.

Bio-inspired decentralized approaches for transporting objects with robot networks seek to use locally-sensed information such as object–robot interaction forces, without the need for robot-to-robot communication. However, the design of the decentralized controller to achieve a specified network performance (e.g., to achieve a desired network settling time $T_{s}$ ) depends on the particular network/object connectivity and therefore, tends to be a centralized decision. Such centralized controller design is not biomimetic and might not be viable if communication is not available between agents to achieve decentralized consensus on the controller parameters. The main contribution of this article is a decentralized controller design approach using local measurements, which does not require prior knowledge of the robot network or object properties. Rather, only the desired network-level performance (such as network settling time) is needed to select controller parameters with the proposed delayed self-reinforcement (DSR) approach, which decentralizes the ideal case where each robot has information about the transport task. In addition, experimental results show that the DSR approach (with decentralized parameter selection) reduces deformation substantially by 66% for a linear object using mobile robots and by 57% for the planar transport of a cylindrical object using industrial robots, when compared to the standard (without DSR) case, even with a centralized design of parameters.

The success of the model predictive path integral control (MPPI) approach depends on the appropriate selection of the input distribution used for sampling. However, it can be challenging to select inputs that satisfy output constraints in dynamic environments. The main contribution of this paper is to propose an output-sampling-based MPPI (o-MPPI), which improves the ability of samples to satisfy output constraints and thereby, increases MPPI efficiency. Comparative simulations and experiments of dynamic autonomous driving of bots around a track are provided to show that the proposed o-MPPI is more efficient and requires substantially (20-times) less number of rollouts and (4-times) smaller prediction horizon when compared with the standard MPPI for similar success rates.

In this work, we present a method for performing ergodic exploration on a region graph using rapid mixing Markov chain. This method enables ergodic planning in spaces with arbitrary topology and at any scale. We have demonstrated in simulation that the ergodic planner outperforms both maximum entropy exploration and random exploration in minimizing the maximum detection error in the context of anomaly detection in a confined space.

Abstract In manufacturing operations such as clamping and drilling of elastic structures, tool–workpiece normality must be maintained, and shear forces minimized to avoid tool or workpiece damage. The challenge is that the combined stiffness of a robot and workpiece, needed to control the robot–workpiece elastic interactions, are often difficult to model and can vary due to geometry changes of the workpiece caused by large deformations and associated pose variations of the robot. The main contribution of this article is an algorithm (i) to learn the robot–workpiece stiffness relationship using a model-free data-based approach and (ii) to use it for applying desired forces and torques on the elastic structure. Moreover, comparative experiments with and without the data-based stiffness estimation show that clamping operating speed is increased by four times when using the stiffness estimation method while interaction forces and torques are kept within acceptable bounds.

This paper proposes an approach to make constant-spacing vehicle platoons robust to large delays and loss of communication. It is well known that centralized communication of the desired trajectory is important to simultaneously guarantee both string stability and constant-spacing in platoons. However, the performance of the resulting connected vehicle system (CVS) is vulnerable to large communication delays and communication loss. The main contribution of this work is a new delayed-self-reinforcement-based (DSR-based) approach that approximates the centralized communication based control by a decentralized predecessor follower (PF) control. The resulting blending of centralized communication with the decentralized DSR approach results in predecessor-leader follower (PLF) control with (i) robustness of the convergence to consensus under large communication delays and (ii) substantially-smaller spacing errors under loss of communication. Comparative simulations show that, for the same level of robustness to internal-stability and string-stability, the variation in settling time to consensus for PLF with DSR under large communication delays is 95% less than PLF without DSR and the steady-state error with DSR under loss of communication is 80% less than PLF without DSR.

The success of the model predictive path integral control (MPPI) approach depends on the appropriate selection of the input distribution used for sampling. However, it can be challenging to select inputs that satisfy output constraints in dynamic environments. The main contribution of this paper is to propose an output-sampling-based MPPI (o-MPPI), which improves the ability of samples to satisfy output constraints and thereby increases MPPI efficiency. Comparative simulations and experiments of dynamic autonomous driving of bots around a track are provided to show that the proposed o-MPPI is more efficient and requires substantially (20-times) less number of rollouts and (4-times) smaller prediction horizon when compared with the standard MPPI for similar success rates. The supporting video for the paper can be found at https://youtu.be/snhlZj3l5CE.