Global Lighthouse Network 2026

Haier’s self-adaptive welding programming agent FIGURE 25 Changeover initiatedWelding parametersgnidleW expert noitalumiS engineer-snamuH in-the loop redael eni LgnidleW expertnoitalumiS engineer ytilauQ inspector ataD analyst gnidleW criticAssigned agents:Prompt library and domain expertise 4 5 Post-pr ocessing 1 Test A Test BSub-task A.2 Simulate welding trajectories using digital twin Sub-task B.2 Adjust parameters based on expert feedbackSub-task B.1 Validate welding programme with simulation engineer Sub-task B.3 Finalize and deploy welding programmeSub-task A.1 Retrieve welding parameters from internal database Sub-task A.3 Optimize welding paths for precision and efficiency2Data and systemsInternal databaseKnowledge base High-pr ecision mapping3D data acquisition 3 slooT (e.g. digital twin, r obotics APIs) Joint-developmentSelf-adaptive welding pr ogramming agent The site faced challenges in meeting the demand for high-end, customized washing machine frames, wher e traditional manual teach-in pr ogramming for welding r obots was time-consuming and inefficient. A r einfor cement lear ning-driven self-adaptive pr ogramming agent was deployed to automate the pr ocess, r educing deviations from 2 to 0.5 millimeters acr oss eight welding r obots. +2.3p.p. dleiy ssap tsriF -45% -nwod ytilauQ grading loss-0.2p.p. -sim edarG classification rateHaier Shanghai, China esitrepxe eht yfidoC of welding engineers and pr ogramming requir ements of high- end washing machine frames gnidlew egagnE engineers and other experts to validate the agents' outputs and provide feedback for improvement etaulave stnegA welding r esults against pr edefined standar ds, and structur ed data models generate executable r obot code ylsuounitnoc stnegA learn from domain experts to self- organize and br eak down tasks into subtasks and execute tnega IA eht elbanE to interact with the real world (i.e. digital twins, r obotics kinematics to simulate 6-axis motion) tpmorP engineering1 dna ataD systems2 slooT 3 slooT 4-tsoP processing5-namuH in-the-loop66 TI dna atad ezinagrO systems to pr ovide context for the AI agent (i.e. sensors and interface pr otocols to collect data)redael eni L namuh A …yb dereenign E stnega I A Source: Global Lighthouse Network. Global Lighthouse Network: Rewiring Operations for Resilience and Impact at Scale 38