Manipulators Control for Grasping Tasks based on Multi-Agent Systems

General Overview

This dataset contains the experimental telemetry and numerical results obtained during the physical validation of a decentralized control architecture for Multi-Agent Robotic Systems (MARS). The data documents the execution of a cooperative caging and coordinated transport task using two 6-degree-of-freedom industrial manipulators (Universal Robots UR3, CB3-Series generation).

This dataset supports the empirical viability of a distributed admittance control scheme that dispenses with external physical force/torque (F/T) transducers, relying instead on indirect force estimation via actuator currents and a first-order IIR digital filter.

Experimental Platform and Methodology

The data were collected on an experimental platform comprising two geometrically facing UR3 manipulators, orchestrated by a computational master node. Cyber-physical communication was established asynchronously via the TCP/IP protocol using Python scripts. During the test, the agents interact with a viscoelastic object made of Thermoplastic Polyurethane (TPU).

The complete experiment (approximately 180 seconds long) evaluates the phases of free approach, contact, caging with a nominal force setpoint ($F_d = 20$ N), transversal transport, and release.

Data Structure

The attached Excel file contains the time series captured during the collaborative manipulation. The main recorded variables include:

Time (t): Timestamp of the asynchronous control cycle (recording network latency and jitter).

Raw and Filtered Force: Readings of the physical interaction on the orthogonal contact axis for both agents ($R_1$ and $R_2$), demonstrating the performance of the stochastic filter with a cutoff frequency of $f_c = 5$ Hz.

Force Setpoint: Constant reference value established to ensure caging without causing plastic deformation of the object.

Admittance Correction : Compensatory Cartesian displacement calculated by the force controller in response to the environmental stiffness.

Positions and Trajectories: Three-dimensional evolution of the end-effectors and the virtual agent (object's centroid) dictated by the Artificial Potential Functions (APF).

Reuse Potential

These data are of particular interest to researchers in collaborative robotics, sensorless force control, and latency compensation in asynchronous cyber-physical networks for industrial manipulation.