{"stable_id":"72fcb19d466f7070","slug":"deepmind-tcv-plasma-control","company":"Swiss Plasma Center at EPFL and Google DeepMind","workflow_name":"Deep-reinforcement-learning magnetic control on the TCV tokamak","function_code":"engineering","pattern_codes":["autonomous_with_backstop"],"changed_assumption":"Operators set plasma objectives while a simulator-trained policy coordinates all coils on physical hardware.","evidence_strength":"verified","publication_tier":"showcase","freshness":"current","reviewed_at":"2026-08-23","updated_at":"2026-08-22","source_quality_summary":"1 peer reviewed; publication outcomes are verified.","caveat_summary":"Research-machine demonstration, not commercial fusion-power operation. Demonstrated configurations do not establish generalization to other tokamaks.","freshness_caveat":null,"workflow_summary":{"before":[{"actor":"Control engineer","action":"Designs a controller for a predefined plasma state and configuration.","actor_type":"human"},{"actor":"Tokamak operator","action":"Runs the engineered controller during an experimental discharge.","actor_type":"human"}],"hinge":"Operators set plasma objectives while a simulator-trained policy coordinates all coils on physical hardware.","after":[{"actor":"Research operator","action":"Specifies plasma shape, current, and location objectives.","actor_type":"human"},{"actor":"RL controller","action":"Commands all magnetic coils at high frequency to track the requested plasma configuration.","actor_type":"system"}],"decision_mode":"bounded_autonomy","decision_marker":"AI acts within a human backstop"},"before":[{"order":1,"actor":"Control engineer","action":"Designs a controller for a predefined plasma state and configuration.","handoff_to":"Tokamak operator","control":"Physics models, engineering constraints, and machine protection."},{"order":2,"actor":"Tokamak operator","action":"Runs the engineered controller during an experimental discharge.","handoff_to":"Research team","control":"Operational safety limits."}],"after":[{"order":1,"actor":"Research operator","action":"Specifies plasma shape, current, and location objectives.","handoff_to":"RL controller","control":"Simulator training plus physical and operational constraints."},{"order":2,"actor":"RL controller","action":"Commands all magnetic coils at high frequency to track the requested plasma configuration.","handoff_to":"Research team","control":"TCV machine protection and operator-defined objectives."}],"decision_rights":"The policy controls coils only within the approved discharge; researchers authorize experiments, objectives, constraints, and shutdowns.","exception_path":"Machine-protection systems and operators terminate or override a discharge when limits are breached.","removed_work":["Configuration-specific low-level controller design for each demonstrated shape"],"outcomes":[{"metric":"Plasma current and shape tracking error","baseline":"A novel learned controller had not previously demonstrated direct full-coil magnetic control on tokamak hardware.","result":"Across the full 0.1-1.0 second interval, RMSE was 0.62 kA for plasma current and 0.75 cm for shape.","period":"Experimental discharges reported in 2022.","scale":"Physical TCV tokamak controlling 19 magnetic coils.","attribution_caveat":"Research-machine demonstration, not commercial fusion-power operation.","evidence_label":"verified"},{"metric":"Configuration range","baseline":"Engineering-driven control of pre-designed states.","result":"The controller produced conventional, elongated, negative-triangularity, snowflake, and sustained two-droplet configurations.","period":"Reported experimental campaign.","scale":"Multiple real-world TCV configurations.","attribution_caveat":"Demonstrated configurations do not establish generalization to other tokamaks.","evidence_label":"verified"}],"executive_lesson":"AI can move humans up from hand-crafting every control law to specifying objectives, but only when simulation, constraints, and machine protection form a hard envelope.","anti_pattern":"Treating a research demonstration as evidence of autonomous commercial fusion operation.","questions_for_leaders":["How faithful is the simulator at edge conditions?","Which shutdown rights remain independent of the model?","What hardware transfer evidence is required?"],"collections":["embodied-work","negative-results"],"bundle_version":"1.0.0","bundle_fingerprint":"sha256:c23c6cc2b88153d008ea8fda928f632ce0011fc2d4c5036672a16e5d895bab93","canonical_url":"https://brianletort.ai/transformations/deepmind-tcv-plasma-control"}