Publications

2024

1. Vector Field Based Adaptive Control for Collaborative Manipulation

   F.B.A. Pessoa and L.C.A. Pimenta

   In Proceedings do XXV Congresso Brasileiro de Automática, 2024

   Abs Bib PDF

   This paper introduces a methodology for computing artificial vector fields designed to ensure the convergence and circulation of trajectories to a curve in the pose space. Extending the work from Rezende et al. (2022) to incorporate orientations, we define a normal vector that asymptotically guides a trajectory to the desired curve, and a tangent vector that ensures circulation. By employing this vector field methodology, we achieve an autonomous closed-loop system with integrated path planning and control. Our approach is applied within a decentralized adaptive control to guide a rigid body with unknown parameters using a team of agents, ensuring compliance with a target pose curve.

   @inproceedings{Pessoa2024,
     author = {Pessoa, Felipe Bartelt de Assis and Pimenta, Luciano Cunha A},
     booktitle = {Proceedings do XXV Congresso Brasileiro de Autom{\'{a}}tica},
     publisher = {SBA Sociedade Brasileira de Autom{\'{a}}tica},
     series = {CBA2024},
     title = {{Vector Field Based Adaptive Control for Collaborative Manipulation}},
     year = {2024},
     keywords = {vector field; adaptive control; collaborative manipulation; motion control; multirobot systems; path planning},
     url = {https://www.sba.org.br/cba2024/papers/paper_7141.pdf},
   }

2025

1. Constructive Vector Fields for Path Following in Matrix Lie Groups

   F.B.A. Pessoa

   Feb 2025

   Abs Bib PDF Video Code Slides

   This work introduces a novel vector field strategy for controlling systems on connected matrix Lie groups, ensuring both convergence to and circulation around a curve defined on the group. Our approach generalizes the framework presented in Rezende et al.(2022) and reduces to it when applied to the Lie group of translations in Euclidean space. Building upon the key properties from Rezende et al.(2022), such as the orthogonality between convergent and circulating components, we extend these results by exploiting additional Lie group properties. Notably, our method ensures that the control input is non-redundant, matching the dimension of the Lie group rather than the potentially larger dimension of the embedding space. This leads to more practical control inputs in certain scenarios. A significant application of this strategy is in controlling systems on SE (3), where the non-redundant input corresponds to the mechanical twist of the object. This makes the method particularly well-suited for controlling systems with both translational and rotational freedom, such as omnidirectional drones. We also present an efficient algorithm for computing the vector field in this context. Furthermore, the strategy is applied as a high-level kinematic controller in a collaborative manipulation task, where six agents manipulate a large object with unknown parameters in the Lie group R3 x SO (3). A low-level dynamic adaptive controller guarantees that the velocity tracking error between the system and the kinematic controller output converges to zero, a result supported by theoretical proofs. Simulation results validate the effectiveness of the proposed method in both the kinematic scenario and the integration of kinematic and dynamic controllers.

   @mastersthesis{pessoa2025constructive,
     author = {Pessoa, Felipe Bartelt Assis},
     title = {{Constructive Vector Fields for Path Following in Matrix Lie Groups}},
     school = {Universidade Federal de Minas Gerais},
     type = {Master's thesis},
     address = {Belo Horizonte, Brazil},
     language = {english},
     year = {2025},
     month = feb,
     keywords = {autonomous systems; guidance navigation and control; tracking; asymptotic stabilization; vector fields; Lie groups; adaptive control; collaborative manipulation},
     url = {https://www.ppgee.ufmg.br/defesas/2189M.PDF},
   }

2022

1. Modelagem e Controle de um Robô para Cirurgias

   F.B.A. Pessoa

   Dec 2022

   Abs Bib PDF Slides

   Surgical robotics is an area, currently, with active development and research. However there are no examples of autonomous surgical systems due to the high complexity and large number of restrictions that govern surgical procedures. Thus, the intention of this work is to implement the da Vinci Si system, well known in the market, in the UAIBot simulator, a simulator developed for didactic purposes by Prof. Vinicius and therefore contains several useful functions for quick implementations of a high range of control algorithms. Thus, it will become possible to implement and develop control strategies in the simulator for the da Vinci. For the implementation of the robot, it became necessary to estimate its Denavit-Hartenberg parameters, since there is no official material with this information, as well as to create collision primitives for distance calculation in environments with obstacles. In order to test the implementation made, it is proposed, by simulation, to control one of the arms of the robot to perform the task of reaching a target pose in an environment with complex obstacles. Given the constraints of the problem, it is of interest to use control algorithms based on quadratic programming that allow minimizing an objective function, given any inequality constraints. Furthermore, it is intended to consider passive joints of the da Vinci Si as actuated, aiming to study the possibility of greater degrees of freedom and, consequently, greater autonomy for the robot. The control strategy, with collision constraints, will be compared with another formulation that does not involve collision constraints. Analyzing the performance of both strategies, it was noted the inability to avoid collisions when these constraints are not explicit. The ability of quadratic programming optimization to converge to a target pose while performing obstacle avoidance was demonstrated. It was observed that the bottleneck for the use of these methods in the context of surgical robotics is the construction of the problem and not its solution, due to the time used to calculate several distances and concatenate all constraints in a single representation.

   @mastersthesis{pessoa2022,
     author = {Pessoa, Felipe Bartelt Assis},
     title = {{Modelagem e Controle de um Robô para Cirurgias}},
     school = {Universidade Federal de Minas Gerais},
     type = {Bachelor's thesis},
     address = {Belo Horizonte, Brazil},
     language = {portuguese},
     year = {2022},
     month = dec,
     keywords = {surgical robotics; da Vinci Si; convex quadratic programming; constrained control; robotic manipulators},
   }