Multi-objective deep deterministic policy gradient for cobot trajectory planning in automotive spot welding

Batyrkhan Sultanovich Omarov

Frontiers in Robotics and AI, 2026

Full text Open access: Gold Metadata (CSL-JSON)
DOI 10.3389/frobt.2026.1939068
Data confidence 70 medium
How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

Introduction Autonomous trajectory planning is a fundamental challenge in collaborative robotic spot welding because positioning accuracy, motion smoothness, operational safety, energy use, and production efficiency must be optimized simultaneously under industrial constraints. Methods This paper presents a Pareto-Guided Multi-Objective Deep Deterministic Policy Gradient (PMO-DDPG) framework for autonomous cobot trajectory planning in automotive spot welding. The framework combines a deterministic actor with four objective-specific critics for accuracy, smoothness, safety, and productivity, together with phase-dependent preferences for the approach, alignment, welding, and retraction stages. The principal novelty is a two-level multi-objective design that preserves six normalized physical reward components, maps them to four specialized critics for phase-dependent actor updates, and uses a non-dominated policy archive for Pareto-guided selection while retaining actor-only online inference. Simulation and industrial experiments assessed convergence, trajectory quality, collision avoidance, energy consumption, cycle time, success rate, and latency. Results Relative to the Soft Actor-Critic (SAC) baseline, PMO-DDPG reduced trajectory error by 66.4%, collision rate by 82.1%, energy consumption by 15.5%, cycle time by 22.0%, and inference latency by 22.7%, while increasing the success rate by 7.5 percentage points. The proposed method achieved 1.08 mm trajectory error, a 1.2% collision rate, 3.65 kJ energy consumption, a 3.87 s cycle time, a 94.6% success rate, and 21.5 ms inference time. Pareto-front analysis produced non-dominated operating points for different manufacturing priorities, and 100 industrial trials confirmed sub-millimeter positioning and repeatable vision-guided welding. Discussion Within the evaluated simulation and industrial test settings, and without a formal significance test, these descriptive results indicate that PMO-DDPG is a promising accurate, safe, and efficient trajectory-planning approach.