Energy-Aware Cooperative Multi-Agent Deep Reinforcement Learning for Sustainable and Precision Trajectory Tracking of an Industrial Robotic Arm: An SDG-Aligned Approach

Control software that lowers actuator effort without degrading precision can be deployed on installed industrial robots without new hardware. This paper formulates trajectory tracking of a 5-degree-of-freedom (DOF) Mitsubishi RV-2AJ arm as a fully cooperative multi-agent problem, in which each joint is driven by a twin delayed deep deterministic policy gradient (TD3) agent under centralized training and decentralized execution (CTDE). Unlike MADDPG, MATD3 and MAPPO, the proposed CTDE-TD3 combines (i) a single twin centralized critic, shared by all joint agents and conditioned on all joint states and torques, which removes the non-stationarity faced by independent agents while each actor uses only three local measurements at run time; (ii) a shared reward with an inter-joint torque-covariance term, whose exact decomposition yields the admissible weight condition μ ≤ 2λ; and (iii) a dimensionless weight normalization. In simulation on three 3D reference trajectories over 10 training seeds, the controller reduced the tracking root-mean-square error (RMSE) by 54.5% and 37.5% and the squared-torque effort by 33.1% and 21.0% relative to centralized and independent-agent TD3, respectively. It was not dominated by MADDPG, MATD3, MAPPO or a tuned proportional–integral–derivative (PID) controller on unseen trajectories. Ablations and a capacity-matched centralized baseline show that the gains stem from the cooperative structure rather than from network size. Because the squared-torque index reflects control effort and copper loss rather than total electrical energy, electricity and CO2 savings are reported only as scenario estimates pending hardware validation.

Authors

Institutions

Publication Details

Journal
Automation
Published
2026-11-18
DOI
https://doi.org/10.3390/automation7060192
Primary Topic
Reinforcement Learning in Robotics
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Energy-Aware Cooperative Multi-Agent Deep Reinforcement Learning for Sustainable and Precision Trajectory Tracking of an Industrial Robotic Arm: An SDG-Aligned Approach

Zied Ben Hazem
4 links
Automation
Reinforcement Learning in Robotics
article

Energy-Aware Cooperative Multi-Agent Deep Reinforcement Learning for Sustainable and Precision Trajectory Tracking of an Industrial Robotic Arm: An SDG-Aligned Approach

Zied Ben Hazem
article en

Abstract

Control software that lowers actuator effort without degrading precision can be deployed on installed industrial robots without new hardware. This paper formulates trajectory tracking of a 5-degree-of-freedom (DOF) Mitsubishi RV-2AJ arm as a fully cooperative multi-agent problem, in which each joint is driven by a twin delayed deep deterministic policy gradient (TD3) agent under centralized training and decentralized execution (CTDE). Unlike MADDPG, MATD3 and MAPPO, the proposed CTDE-TD3 combines (i) a single twin centralized critic, shared by all joint agents and conditioned on all joint states and torques, which removes the non-stationarity faced by independent agents while each actor uses only three local measurements at run time; (ii) a shared reward with an inter-joint torque-covariance term, whose exact decomposition yields the admissible weight condition μ ≤ 2λ; and (iii) a dimensionless weight normalization. In simulation on three 3D reference trajectories over 10 training seeds, the controller reduced the tracking root-mean-square error (RMSE) by 54.5% and 37.5% and the squared-torque effort by 33.1% and 21.0% relative to centralized and independent-agent TD3, respectively. It was not dominated by MADDPG, MATD3, MAPPO or a tuned proportional–integral–derivative (PID) controller on unseen trajectories. Ablations and a capacity-matched centralized baseline show that the gains stem from the cooperative structure rather than from network size. Because the squared-torque index reflects control effort and copper loss rather than total electrical energy, electricity and CO2 savings are reported only as scenario estimates pending hardware validation.

AutomationVol. 7(6)
University of Technology Bahrain (BH)
Industry, innovation and infrastructure
Reinforcement Learning in Robotics
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon