Singularity-Free Guiding Vector Field over Bézier’s Curves applied to Rovers Path Planning and Path following.

Alfredo González-Calvin, Lía García-Pérez, Juan F. Jiménez · 2024

Read Singularity-Free Guiding Vector Field over Bézier’s Curves applied to Rovers Path Planning and Path following. by Alfredo González-Calvin, Lía García-Pérez, Juan F. Jiménez (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-13033.txt.

Introduction

Alfredo González-Calvin, Lía García-Pérez, and Juan Jiménez present a guidance method for wheeled robots in this 2024 paper. The robots must follow a set path on the ground while respecting limits such as turning radius and side force. Their method builds a field of guide directions that the robot can follow toward the path.

The new field avoids the dead spots that older fields can have, by working in a higher space. Paths are drawn as smooth Bézier curves, which fit the method well. A speed rule that varies with curve sharpness helps the robot settle onto the path. The team gives theory, computer tests, and outdoor runs with a robot built from shop-bought parts.

Catalog record

CREATORAlfredo González-Calvin, Lía García-Pérez, Juan F. Jiménez
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE9 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-13033

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Singularity-Free Guiding Vector Field over Bézier’s Curves applied to Rovers Path Planning and Path following. Alfredo González-Calvin, Lía García-Pérez, Juan F. Jiménez Abstract This paper presents a guidance algorithm for solving the problem of following parametric paths, as well as a curvature-varying speed setpoint for land-based car-type wheeled mobile robots (WMRs). The guidance algorithm relies on Singularity-Free Guiding Vector Fields SF-GVF. This novel GVF approach expands the desired robot path and the Guiding vector field to a higher dimensional space, in which an angular control function can be found to ensure global asymptotic convergence to the desired parametric path while avoiding field singularities. In SF-GVF, paths should follow a parametric definition. This feature makes using Bezier’s curves attractive to define the robot’s desired patch. The curvature-varying speed setpoint, combined with the guidance algorithm, eases the convergence to the path when physical restrictions exist, such as minimal turning radius or maximal lateral acceleration. We provide theoretical results, simulations, and outdoor experiments using a WMR platform assembled with off-the-shelf components. Keywords Wheeled Mobile Robots, Guiding Vector Fields, Parametric Paths, Path following, Speed controller, curvature changing speed setpoint, Rover. 1 Speed Control Problem Several works can be found in the literature on adapting guidance controllers to the curvature, like in [ Leng and Minor, 2017 ] , for guidance algorithms to achieve better convergence. A speed controller for an AMR vehicle is helpful when following curvature-changing paths. This allows us to reach a constant speed setpoint and adapt the speed to path characteristics. The non-varying setpoint provides for constant speed path-following while changing the setpoint depending on some information about the desired path (e.g., curvature) should speed up the convergence of the guidance controller. This approach is simple because only the desired reference speed changes; thus, whatever speed controller can be used, for example, in [ Cao et al., 2019 ] , a finite time double sliding surface guidance algorithm is used for Subway’s speed curve tracking. In recent years, there has been increased attention towards determining trajectory-dependent speed profiles for guidance due to the rise of autonomous road vehicles [ Gámez Serna and Ruichek, 2017 ] . Most works use a-priori information about the desired trajectory to compute the path curvature and use this information to adapt vehicle speed, preventing slipping and rollover. Trajectory information can be extracted from maps, like [ Villagra et al., 2012 ] that used a closed-form speed profiler combined with the path planner, providing a continuous velocity reference. Or [ Li et al., 2012 ] that obtains the curvature using the Geographic Information Systems (GIS) roadmaps. In other cases, certain assumptions are made about the curvature of roads to adapt the vehicle’s speed, as in [ Park et al., 2015 ] . 1.1 Speed controller To control the Rover speed, we use as a control signal u_{v} (i.e., the throttle) the output of a feedforward + PID controller: u_{v}=k_{f}v_{ref}+k_{p}e_{v}(t)+k_{d}\dot{e}_{v}(t)+k_{i}\int_{0}^{t}e_{v}(\tau)d\tau, (1) where v_{ref}\in\mathbb{R}_{\geq 0} is the desired speed, e_{v}(t)=v_{ref}-v(t) is the error between this one and the actual speed of the Rover, and k_{f},k_{p},k_{d},k_{i}\in\mathbb{R}_{\geq 0} are, respectively, the feedforward, proportional, derivative and integral control constants. The feedforward controller allows reaching a constant speed depending on v_{ref} , while the PID controller acts in the presence of error, allowing the desired speed to be reached without steady-state error. Recall that, as was pointed out in section , we have ignored the dynamic of the Rover. Thus v\sim\propto u_{v} and we can consider, v=k_{f}v_{ref}+k_{p}(v_{ref}-v)+k_{d}(\dot{v}_{ref}-\dot{v})+k_{i}\int_{0}^{t}(v_{ref}-v)dt (2) Taking time derivatives, considering a constant v_{ref} and regrouping terms, we get a linear system, \begin{pmatrix}\ddot{v}\\ \dot{v}\end{pmatrix}=\begin{pmatrix}-\frac{1+k_{p}}{k_{d}}&-\frac{k_{i}}{k_{d}}\\ 1&0\end{pmatrix}\begin{pmatrix}\dot{v}\\ v\end{pmatrix}+\begin{pmatrix}1\\ 0\end{pmatrix}\frac{k_{i}}{k_{d}}v_{ref}. (3) It is immediate to see that the system matrix always has negative spectral abscissa, i.e. the system is stable and v\rightarrow v_{ref} as t\rightarrow\infty , due to the control integral term. Note that if the setpoint change is slower than the controller’s time constant, then the vehicle will follow the desired time-changing speed setpoint. This plays a crucial role in the next section since we will use a smooth (i.e., differentiable) and variable setpoint that depends on curvature. In our case, the Rover measures its position directly from the GPS, discarding data from inertial sensors. Our Rovers are small outdoor vehicles subject to bumpy and stony terrain and several other factors. This bounciness affects the accelerometers of the IMU, adding a significant amount of noise, almost equal to or larger than the actual acceleration of the Rover due to displacement. Besides, the Rover’s acceleration transient time is faster than the noise perturbations (due to its relatively small mass). Therefore, the acceleration due to an actual increase or decrease in speed could not be differentiated from the noise signal. In addition, a proper measure of the noise process variance is complex since it depends on the terrain. Thus, the GPS speed is critical, but this signal is also noisy. Filtering is needed to allow for the operation of a speed controller (especially the derivative controller). For this purpose, we use a simple moving average filter. Therefore, our Rovers measure the speed using only the GPS signal filtered with a moving average filter, and they use this filtered speed measurement v as an input to the speed controller in equation ( 1 ). 1.2 Changing speed setpoint Typically, path-following algorithms consider a constant speed when following a specific curve. This speed is chosen so that the vehicle can follow the trajectory with sufficiently small (guidance) control inputs. Moreover, their heading cannot change instantly when dealing with non-holonomic vehicles like the model in equation (1). This means that the radius of curvature that it can trace increases (respectively decreases) as speed increases (decreases). Therefore, if the speed is held constant, it must be restricted to the maximum value that ensures the Rover can follow the trajectory. That is, the speed of the Rover must allow its guidance controller to converge to the trajectory, regardless of high curvature segments. That is because guidance control signals (i.e., inputs to the steering angle) at relatively high speeds can cause a rollover. Moreover, the saturation of the steering angle creates the same problems explained above. Besides, there are applications in which a constant speed is not desirable or interesting. For example, if the trajectory must be completed below a certain time stamp, an interesting approach is that the vehicle accelerates where the curvature is low or zero (i.e., straight lines) and reduces its speed when curvature increases. For this purpose, we propose that the speed setpoint can be changed according to the path’s curvature, following the next equation, v_{ref}(\kappa)=(v_{max}-v_{min})\exp({-c_{\kappa}\kappa^{2}})+v_{min}, (4) where v_{max} and v_{min}\in\mathbb{R}_{\geq 0} are the maximal and minimal Rover speed ( v_{max}\geq v_{min} ), \kappa\in\mathbb{R} is the curvature and c_{\kappa}\in\mathbb{R}_{\geq 0} is a constant used to penalize the speed as curvature increases. Since we are dealing with parametric curves, the curvature formula is well known [ Dubrovin et al., 1984 ] and is defined as \kappa(w)=\frac{f_{1}^{\prime}(w)f_{2}^{\prime\prime}(w)-f_{1}^{\prime\prime}(w)f_{2}^{\prime}(w)}{(f_{1}^{\prime}(w)^{2}+f_{2}^{\prime}(w)^{2})^{3/2}}. (5) Thus, the curvature is computed using the parameter of the path w , that is, equation ( 5 ) calculates the curvature of the point at which the vector field is pointing to (see figure ). Assuming that c_{\kappa} is greater than zero in equation ( 4 ), we can observe that

Read the complete text