Session 7B – Robot Manipulation and Control
Date: Friday, 20th November 2026
Time: 09:00 – 10:30
Room: Aula Capella
Session Chair: TBA
Presentation format: Each paper is allocated 15 minutes: approximately 10–12 minutes for the presentation and 3–5 minutes for questions.
09:00 – Robotic Arm with Motion Control through Computational Vision
Authors: Henrique Abrantes¹, Christian Rodrigues¹, Rodrigo Maria¹, Pedro Teodoro², Ricardo Póvoa³, Camilo Christo¹, Gonçalo Piedade¹, Paulo Luz¹ and Anna Finamore¹
Affiliations: ¹ Escola Sup. Náutica Inf. D. Henrique, Paço de Arcos, Portugal; ² MARE-Instituto Politécnico de Setúbal and Escola Sup. Náutica Inf. D. Henrique, Paço de Arcos, Portugal; ³ IT-Instituto de Telecomunicações and Escola Sup. Náutica Inf. D. Henrique, Paço de Arcos, Portugal.
Abstract
This paper focuses on the development of a humanoid robotic arm, controlled by a computer vision system. It presents the design and assembling of a functional prototype with 2 Degrees of Freedom (DoFs): base and wrist rotations; and an anthropomorphic hand with 5 independent fingers, adding another 5 DoFs to the system. The control is based on real-time image processing, where the proposed system interprets human movements and replicates them in real-time using servomotors. Regarding the mechanics, a 22 Factor of Safety (FoS) structure was designed and produced with additive manufacturing. This work displays the integration of 3D additive manufacturing, distributed control electronics, computer vision and image processing. The results show a stable processing rate of around 20 Frames Per Second (FPS) and, in qualitative tests, the ability to mimic the movement of a human upper limb, supporting open-source, intuitive and low-cost anthropomorphic-based human-machine interfaces built with accessible technologies such as Raspberry Pi and MediaPipe. An innovative algorithm for real-time and dynamic axis switch is also proposed. The extension to full-body prototyping is envisaged as future work.
09:15 – Active Exploration in Bayesian Model-based Reinforcement Learning for Robot Manipulation
Authors: Carlos Plou, Ana C. Murillo and Ruben Martinez-Cantin
Affiliations: DIIS-I3A, University of Zaragoza, Zaragoza, Spain.
Abstract
Efficiently tackling multiple tasks within complex environments, such as those found in robot manipulation, remains an ongoing challenge in robotics and an opportunity for data-driven solutions such as reinforcement learning (RL). The high cost of data gathering in robotics encourages data-efficient approaches such as model-based RL, which builds a dynamic model of the robot to conduct most policy learning on cheaper simulations, while enabling data reuse and transfer between tasks. In this work, we employ Bayesian deep learning (BDL) with a double purpose. First, we increase the model quality by improving its calibration. Second, we improve data efficiency by performing active learning of the dynamic model during a preliminary exploration phase, actively estimating the information gain of each transition and using it as the exploration reward. We compare several BDL methods in a realistic robot manipulation setup. Our experiments show that our approach matches the quality of relevant alternatives with much lower robot execution requirements and, unlike previous studies validated only on standard RL environments, we tackle robotic arm end-tasks on a complete manipulator simulator such as CoppeliaSim. Code is available at ropertunizar/BayesianActiveExploration .
09:30 – Evaluation of Wave-Based Teleoperation for Intuitive Collaborative Robot Manipulation in Industrial Environments
Authors: Mona Saber Gharamaleki, Mattia Tafuri, Claudio Melchiorri, Gianluca Palli and Roberto Meattini
Affiliations: Department of Electrical, Electronic, and Information Engineering, University of Bologna, Bologna, Italy.
Abstract
Bilateral teleoperation enables a human operator to control a remote robotic system while receiving haptic feedback from the environment. Although wave variable transformations are a well-established method for ensuring stability in the presence of communication delays, their application to collaborative robots in industrial settings remains limited. In this work, we present a bilateral teleoperation architecture based on a wave-variable transformation of the exchanged force and velocity signals, together with dynamic and user-configurable scaling. The approach leverages the inherent passivity of the wave channel while exploiting its underlying dynamic structure to enable intuitive interaction in both precision and high-mobility tasks. We demonstrate the implementation of the proposed scheme on a real-world setup consisting of a 7-DOF collaborative manipulator and a stylus-based haptic interface, evaluating performance through both simulation and experimental trials. An experimental evaluation based on a weight-discrimination task shows that the force–velocity modulation supports task execution and haptic perception in a preliminary single-subject pilot study (N = 1), while stable closed-loop behavior is observed under the tested conditions. The findings demonstrate the feasibility of wave-based teleoperation with scalable feedback in the evaluated collaborative-robot scenario and motivate further investigation toward industrial teleoperation applications.
09:45 – Adapting to Variable Active-Joint Configurations via Kinematics-Informed Attention for Robot Arms
Authors: Patrick Thomas Mayer and Jan Rexilius
Affiliations: Bielefeld University of Applied Sciences and Arts, Minden, Germany.
Abstract
Deep reinforcement learning (RL) policies for robotic control are typically trained for a single hardware configuration. When a joint locks, the learned mapping no longer matches the available actuation. This paper investigates whether combining self-attention with explicit kinematic observations can improve a policy’s adaptability to varying active-joint configurations. We train a single RL agent to control a Franka Emika Panda manipulator across different sets of active joints, from a restricted 4-degree-of-freedom (DOF) mode to the fully redundant 7-DOF configuration. Instead of fixed-length state vectors, the method processes the active joints as a variable-length sequence, enriching each joint’s representation with its relative spatial offset and geometric Jacobian influence. The proposed kinematics-informed attention architecture reaches a success rate of 82.1 % across trained configurations and 66.6 % on unseen configurations, outperforming multilayer perceptron (MLP) and generic attention baselines while remaining nearly collision-free. Deployed on a physical Panda, the simulation-trained agent reaches 79.0 % success on the unseen configurations, closely matching simulation on the same target schedule (77.0 %).
10:00 – VAMPiric ROS 2: A Framework for Real-Time Dual-Arm Motion Planning and Execution
Authors: David Yu Wu and Jan Rosell
Affiliations: Institute of Industrial and Control Engineering (IOC), Universitat Politècnica de Catalunya (UPC), Barcelona, Spain.
Abstract
Real-time dual-arm manipulation in dynamic environments requires fast motion planning, efficient environment representation, and seamless integration with robotic platforms. This paper presents VAMPiric ROS 2, a robot-agnostic ROS 2 framework that integrates the Vector-Accelerated Motion Planning (VAMP) into a complete dual-arm manipulation pipeline. The framework supports independent coordinated planning for each manipulator, inter-arm collision avoidance, robot-body and environment representation, trajectory generation, and execution within a unified ROS 2 architecture. Point-cloud obstacles are incorporated using Collision-Affording Point Trees (CAPT). The proposed architecture was validated through real-world experiments on the MADAR platform and simulation using the TIAGo dual-arm robot. Experimental results demonstrate the feasibility of continuous real-time replanning while integrating perception, collision avoidance, and coordinated dual-arm motion within a unified framework.
10:15 – Single-Controller Teleoperation and Multimodal Dataset Generation for a Mobile Manipulator
Authors: Fardien Azizi¹, Sitor Sla¹, Christiaan Peppelman¹, Dion van Santen¹, Andrei-Carlo Papuc¹, Khaled Mustafa¹, Diego Martinez-Baselga¹,² and Javier Alonso-Mora¹,
Affiliations: ¹ Department of Cognitive Robotics, TU Delft, Delft, The Netherlands; ² RoPeRt Group, DIIS-I3A, University of Zaragoza, Zaragoza, Spain. The first four authors contributed equally.
Abstract
Robot learning systems require synchronized, high-quality demonstrations that combine visual, proprioceptive, and action data. However, collecting such datasets with mobile manipulators is often slowed down by fragmented teleoperation interfaces and manual post-processing workflows. This paper presents an integrated framework for teleoperated multimodal data collection on a Clearpath Dingo-O mobile base equipped with a Kinova Gen3 Lite manipulator and an Intel RealSense RGB-D camera. The framework combines single-controller operation of the base, arm, and recording process with an automated ROS-based pipeline for data capture, image-quality filtering, temporal synchronization, visualization, and export to machine-learning-ready Zarr datasets. We validate the system on a pick-and-place mobile-manipulation task in which the robot transports a bottle between two surfaces. The results show that the unified controller supports complete task execution and recording without external commands, while the data pipeline produces synchronized datasets for short demonstrations and identifies corrupted visual frames. The proposed framework provides a practical basis for producing structured demonstrations for robot-learning research. Code and videos of the experiments can be found at this repo .