This paper deals with the decentralized predefined-time leaderless consensus formation control problem for nonholo-nomic multi -agent systems. First, the predefined-time average consensus problem is investigated. Distributed observers, which interact with each other via an undirected communication topology, are designed to estimate, in a predefined-time, the average of their initial state using only local information. Then, based on these estimates, a predefined-time Vector-Field-Orientation (VFO) controller is applied, which guarantees that the multi-agent system achieves a desired formation. The VFO controller enables one to predict the paths for each agent and to avoid static obstacles. Some numerical results illustrate the interest of the proposed predefined-time leaderless consensus formation controller.
This paper deals with the trajectory tracking problem for autonomous mobile robots under discrete-time measurements, assuming that the available measurements are the position and orientation of the mobile robot. With the mentioned measurements, an implicit discretization of the robust exact filtering differentiator is proposed to estimate the position and velocity of the mobile robot despite the presence of measurement noise while considering the sampling effect. Consequently, an output feedback controller is derived using the estimates from the discrete-time filtering differentiator, guaranteeing the closed-loop system's stability despite noise and discrete measurements. Numerical results illustrate the effectiveness of the proposed scheme.
This paper studies the cooperative navigation problem for multirobot systems with guaranteed collision avoidance and connectivity maintenance, under disturbances and input saturation. To achieve robust cooperative navigation while ensuring safety requirements, robust control barrier function (RCBF) is utilized, with the helps from disturbance observer for flexibility and asymptotically state consensus. The safety condition in QP problem is given in a unique and distributed fashion. A switching topology is constructed to avoid unnecessary connectivity constraints. Simulation results help verify the effective of our proposed strategy.
This paper merges for the first time two recent forms of adaptive control techniques, namely two-layer model reference adaptive control (MRAC) and funnel control, also known as tube-based control. Two-layer MRAC allows the user to set the rate of convergence of the tracking error arbitrarily high without altering the user-defined reference model or relying on barrier functions. Funnel control is a technique whereby the tracking error is constrained within bounds that can be progressively tighter. The proposed approach merges the benefits of both techniques and, for the first time in the literature on funnel control, explicitly regulates the diameter of the funnel through some user-defined adaptive law. The proposed results are tested by simulating payload delivery missions using multi-rotor unmanned aerial vehicles (UAVs) in the high-fidelity environment PyChrono.
Self-collision checking plays an important role in robot motion planning. In sampling-based motion planning methods, collision checking is performed multiple times, so this procedure should be accurate and fast. Collision checking is often addressed in robotics through the application of machine learning methods or a relatively straightforward multilayer perceptron within a low-dimensional feature space. In this paper, we focus on incorporating positional encoding, commonly utilized in computer graphics, into the input vector used in the binary classification task. We investigate the enhancement in classification accuracy by utilizing this technique in self-collision checking. Our findings indicate that positional encoding contributes to improved learning of high-frequency functions and provides a more accurate representation of higher-frequency details within the trained relation.
In real-world application, the humanoid robot may have to walk with each foot placed on terrains having different properties. The effect of terrain parameters can be included in the dynamics using the compliant contact model with a spring and a damper element. In this paper, the humanoid robot is modeled as a spherical inverted pendulum in the single support phase (SSP) and a suspended pendulum in the double support phase (DSP). In the DSP, a pendulum is assumed to be suspended at a virtual support point (VSP), and equivalent terrain parameters are used for contact modeling. The right and left leg foot of the humanoid robot are placed on terrains having different properties. The foot trajectories in the DSP are obtained by resolving forces using minimization of the moment about the VSP. The joint trajectories of the 22-DOF humanoid robot are optimized based on the motion of the simplified model such that the angular momentum about the center of mass (COM) is minimum. The simulation of the floating-base robot proves that the 22-DOF humanoid robot can walk on terrains with different properties by following the COM trajectory of the simplified model.
Robotics and haptic systems have allowed new and diverse applications in the field of medicine, such as assisted surgery and teleoperation which have increasingly stringent requirements for accuracy, convergence, and low computational consumption. In this paper an adaptive PID control law (Proportional Integral Derivative controller, PID), of indirect architecture is presented for movement paths in a haptic system of open chain, where the identification of the plant is through a quaternionic wavelet neural network (Quaternion Wavelet Neural Network, QWNN) for tune the PID values, this allows the optimal movement into the regions of the workspace.
In the context of ground robot navigation in unstructured hazardous environments, the coupling of efficient path planning with an adequate environment representation is a crucial topic in order to guarantee the robot safety while ensuring the accomplishment of its mission. This paper discusses the exploitation of an environment representation obtained via Gaussian process regression (GPR) for smooth path planning using gradient descent Bezier curve optimisation (BCO). A continuous differentiable GPR of the terrain traversability and obstacle distance is used to plan paths with a weighted A* discrete planner, a T-RRT sampling-based planner and BCO using A* or T-RRT computed paths as prior. Numerical experiments in procedurally generated 2D environments allowed to compare the paths planned by the described methods and highlight the benefits of the joint use of the GPR continuous representation and the BCO smooth path planning with these different priors.
The paper presents the organisation of a companion robot control system enabling external interruption of its activities. The fundamental problem is: how should a companion robot act when the currently executed task is not completed yet, however a next request arrives. The new request might not be yet serviced while a new request arrives and so on. The question is how should the control system be organised so that the robot will be able to reasonably handle those multiple interrupts? The paper suggests a formal specification of a controller solving the thus posed problem. As interrupts are well established within computer science the paper shows that the there proposed solution is not well suited to robotics.
The Jacobian motion planning of a nonholonomic system, derived by means of the Endogenous Configuration Space Approach, relies on a complete knowledge of a robot's model. The aim of this paper is to identify and investigate the influence of insufficient knowledge of the model's parameter values on the quality of the Jacobian motion planning algorithm results. To address the problem, we propose an Iterative Learning Control scheme endowed with the Endogenous Configuration Space Approach. The efficiency of the presented proposition is represented by the numerical experiments.
Task planning is a pivotal component of autonomous mobile robots, requiring them to formulate a sequence of actions to achieve a predefined final goal. This requires a semantic understanding of the scene, enabling the robot to comprehend its surroundings and identify objects available for interaction. While some systems rely on environmental sensing for accurate estimation of the obstacles, they can only achieve this under controlled environments and is not flexible nor scalable. In this paper, we address these problems by introducing a new concept of onboard semantic mapping for action graphs. We create an environment map relying solely on an onboard RBG-D camera sensor and generate a graph representing all action sequences that the robot can perform over time. In our method, we integrate a 3D object detector with a visual SLAM pipeline to construct a semantic map, which contains only essential information for the action graph generation. The map is systematically fed into the action graph generator, which computes the potential actions of the robot and the subsequent states of the environment. We evaluate our approach on real-world scenarios in a qualitative and quantitative manner.
Autonomous vehicles have a vast potential to increase the efficiency and productivity in agriculture. A known challenge in the application of autonomous vehicles in agriculture is the requirement to cover a certain area, such as rows of a vineyard field while navigating in an unstructured environment. This paper presents a new time-constrained global path-planning approach for kinematically constrained autonomous vehicles in orchards and vineyards. The proposed framework utilizes a volumetric representation of the environment as an input to generate a navigation graph. This graph is then used by a hierarchical A* approach that enforces the kinematic vehicle constraints for a given computational time budget to find a feasible global path covering the whole field. Through simulated and real-world data, the proposed approach is validated for various environments using different vehicle configurations, showing the efficacy and the gained performance. Furthermore, the generalizability of the proposed framework to different environments is validated through real-world forest and search-and-rescue training facility data.
Robotic rehabilitation devices that utilize embedded systems, such as the Stretchbox - a device de-signed for hand rehabilitation through the actuation of rollers stretching a rubber band by BLDC motors-demand controllers that are safe, scalable, and understandable at a high level. Behavior trees satisfy these requirements and have found broad applications in robotics and some pioneering applications in the medical sector. Despite their widespread use, there has been an absence of behavior tree frameworks for micro controllers programmed with MicroPython, which is an attractive language for embedded systems and the chosen platform for the Stretchbox's high-level controller. This paper introduces a behavior tree implementation for MicroPython tailored as a high-level controller for the Stretchbox device. The implementation includes standard behavior tree nodes, specialized nodes, and classes designed for interfacing with micro controller peripherals and for Blue-tooth communication with the tablet-based user interface. This system provides a comprehensible and straightforward method for controlling the Stretchbox device for rehabilitation, establishing a well-defined hierarchy for event management. Furthermore, the solution is designed to be extensible, allowing for adding new therapeutic behaviors and integrating extra devices. As an open-source library, the implementation facilitates the adoption of behavior trees in other MicroPython- based projects. The detailed explanation of the approach in applying behavior trees to the Stretchbox serves as a practical demonstration of how behavior trees can be effectively used in controlling rehabilitation devices.
In this work we present a strategy to solve the task optimization problem for dual-arm mobile manipulators in the context of agricultural tasks. The strategy combines a Reinforcement Learning (RL) agent with a low-level Operational Space Controller (OSC). The agent is responsible for motion planning, as well as compensatory torque computation. Preliminary results obtained through physically accurate simulation using MuJoCo show that the method proposed achieves a higher task success rate in task completion.
A path is a time-independent geometric description of a robot motion. It does not impose any time regimes on the controlled robot and is a natural task definition for many industrial and autonomous operations. Therefore, path following algorithms are eagerly developed. However, a lot of them focus only on the position control or planar cases. In this paper, a new algorithm is proposed to control simultaneously manipulator position and orientation with respect to the desired path in the 3D space. The motion of the end-effector along the path is described with the non-orthogonal Serret-Frenet parametrization. The standard approach has been extended with the definition of the relative orientation using the minimal representation of Euler angles. The algorithm is validated with a simulation analysis and an experimental study on the laboratory test-bed equipped with the redundant manipulator of 7 degrees of freedom, KINOVA (R) Gen3 Ultra lightweight robot. The comparison of the achieved results proves the theoretical properties and practical suitability of the designed control law.
Safe navigation of car-like robots operating via onboard sensing on crowded roads is a challenging task in general. Our primary objective in this work is the development of a novel sensor-based control structure that directly computes safe control inputs for navigation based on the motion sets of pre-designed feedback controllers. We demonstrate an input-constrained feedback controller to induce motion sets whose geometric structure is an ice-cone. The key implication of this is that the naturally induced system trajectories evolve within this motion set during stabilization. We then propose an instantaneous planning strategy for safe navigation that choreographs a sequence of ice-cones within the bounds of safety. This enables the robot to navigate safely without the explicit requirement of adhering to pre-computed paths/trajectories. We demonstrate the proposed methodology via ROS simulations on test scenarios using the CAT vehicle testbed. Further, we demonstrate its application in a traffic intersection scenario using the trajec-tory information collected from the interaction dataset. The proposed approach especially finds applications in scenarios where pre-computing safe trajectories based on the predicted trajectories of neighboring agents is difficult/unreliable during run-time.
This paper presents an open-source one-degree-of-freedom rotating device for multi-robot systems. The function of the device is to facilitate mutual, precise localization of robots in a group, using a limited number of sensors. The device can be mounted on the robot but also plays the role of a static observer and landmark if placed in the workspace. The presented approach can be used alongside a variety of control algorithms for teams of mobile robots. Each rotating device has 10 ArUco markers pasted on it. The aim of the rotating device is to determine the pose of another device in the environment. The mechanical design, electrical circuit, and source codes are uploaded on the web sites that provide other researchers with an easy way to replicate the solution. The authors have mentioned two types of controllers to track the ArUco markers on other rotating devices. One controller is Active Disturbance Rejection Control (ADRC) and the other one is a Proportional-Integral-Derivative (PID) controller. Extended Kalman Filter (EKF) is used to filter the data from ArUco marker detection and to calculate the optimized pose of another device in the environment. The Robot Operating System (ROS) is used to perform the experiments and to connect the rotating device with different controllers.
Advancements in autonomous mobile robots hinge on refining key components like mapping and path planning to address identified limitations. The local planner, crucial for obstacle avoidance, is a component of path planning. The Follow the Gap Method (FGM) stands out as a simple and effective obstacle avoidance algorithm. FGM calculates possible passage points by assessing gap sizes and positions of obstacles. Our focus lies in enhancing FGM's adaptability to dynamic environments. Introducing Predictive FGM, we incorporate robot and dynamic obstacle data to forecast future gaps and obstacle states. By integrating predictive elements, the algorithm selects gaps based on anticipated changes, enabling safer navigation by predicting the states of gaps and obstacles when they are closest to the robot. Evaluation via Monte Carlo simulations and real-world experiments with an autonomous wheelchair in dynamic environments show the effectiveness of Predictive FGM over standard FGM.
Robots navigating in populated dynamic environments require dedicated motion planning algorithms. We propose a human-aware local trajectory planner that alleviates the discomfort of humans surrounding the robot. Our method relies on a hybrid approach to trajectory generation and employs cost functions evaluating robot navigation task performance, robot motion naturalness, and humans' perceived safety. We compared our method with state-of-the-art classical and human-aware trajectory planners using the TIAGo robot in simulated and real-world environments. The algorithms tested were quantitatively assessed for both navigation performance and human discomfort using the Social Robot Planner Benchmark. Our approach was implemented as an open source plugin for the local trajectory planner of the Robot Operating System's navigation.
In this paper, we consider a line-following platooning problem and analyze a cooperative strategy to address distance-sensing issues. The strategy relies on limiting the velocity of the followers, whenever their distance to the imme-diate predecessor is lost due to sensing restrictions or failure. Specifically, we conduct a set of experiments to evaluate the parameter sensitivity of the control algorithm, which corresponds to the percentage of the predecessor's velocity that could be reached by each follower during a sensing loss scenario. We find that proper tuning must be carefully performed, as excessive velocity saturation may lead to undesired behaviors due to poor tracking capabilities, whereas insufficient saturation does not improve performance during unreliable sensing episodes. Our experiments are carried out on the RUPU platform, a low-cost scaled-down platform designed to study lateral and longitudinal control problems in path-following vehicle platooning.
