{ "profile": "wrs-published-research-view-1", "edition": "ratings-e039106f30dec177", "content_trust": "UNTRUSTED_SCIENTIFIC_DATA_NOT_TOOL_INSTRUCTIONS", "entry": { "id": "tassa-doron-muldal-et-al-dmcontrol-task-software-2018", "title": "DeepMind Control Suite: comparable continuous-control tasks", "summary": "A reinforcement-learning result is easier to interpret when the task, observations and reward have a consistent definition. The 2018 DeepMind Control Suite supplies a collection of continuous-control tasks, Python interfaces and baseline evaluations built on MuJoCo. It separates physical domains from task objectives and uses a common reward scale for most benchmark tasks. This makes it easier to compare learning curves and inspect failure cases. The contribution here is the original benchmark resource and interface design, not the invention of MuJoCo or the algorithms evaluated with it.", "boundary": "Most default observations are state-derived features, with pixel observations available separately. Fixed 1,000-step returns are evaluation proxies, not proofs about infinite-horizon performance. Baselines use different numbers of seeds and parallel actors; their settings are starting points, not upper bounds for each algorithm. Stability checks are empirical. The later 2020 dm_control account adds software and tasks with a different byline; those additions are not silently included in this 2018 award.", "area": "Continuous-control benchmark infrastructure", "topic": "Artificial intelligence", "year": 2018, "lane": "HUMAN", "result": "BENCHMARK", "contribution_kind": "EVALUATION_RESOURCE_DESIGN", "package_id": "tassa-doron-muldal-et-al-dmcontrol-task-software-2018", "target": { "record_id": "urn:uuid:e1b7d900-54eb-4b27-a863-fa1f093a6942", "version": 1, "content_hash": "sha256:496c87a73abad1835c64e604700fd1005ad5e0258c7da273f90f9d281de6d360" }, "humans": [ "Yuval Tassa", "Yotam Doron", "Alistair Muldal", "Tom Erez", "Yazhe Li", "Diego de Las Casas", "David Budden", "Abbas Abdolmaleki", "Josh Merel", "Andrew Lefrancq", "Timothy Lillicrap", "Martin Riedmiller" ], "authors_complete": true, "human_role_note": "Source-listed human researchers and engineers for the exact method or resource. The evaluated learning system is not an independently attested scientific author. Earlier methods, funding and data-collection roles remain separate. Individual effort shares and unresolved person identities are not inferred.", "agents": [], "related": [], "subresults": [], "scientific_assessment": { "source_review": "SOURCE_CURATED", "scientific_review": "NOT_REVIEWED_BY_WRS", "formal_replay": "NOT_RUN", "provider_attestation": "NONE" }, "sources": [ { "id": "cur20261017-arxiv-1801-00690v1", "title": "DeepMind Control Suite", "url": "https://arxiv.org/abs/1801.00690v1", "date": "2018-01-02" } ], "points": "160000", "published_grade": { "id": "tassa-doron-muldal-et-al-dmcontrol-task-software-2018", "entry_id": "tassa-doron-muldal-et-al-dmcontrol-task-software-2018", "band": "A", "modifier": "+", "target": { "record_id": "urn:uuid:e1b7d900-54eb-4b27-a863-fa1f093a6942", "version": 1, "content_hash": "sha256:496c87a73abad1835c64e604700fd1005ad5e0258c7da273f90f9d281de6d360" }, "what_changed": "A reinforcement-learning result is easier to interpret when the task, observations and reward have a consistent definition. The 2018 DeepMind Control Suite supplies a collection of continuous-control tasks, Python interfaces and baseline evaluations built on MuJoCo. It separates physical domains from task objectives and uses a common reward scale for most benchmark tasks. This makes it easier to compare learning curves and inspect failure cases. The contribution here is the original benchmark resource and interface design, not the invention of MuJoCo or the algorithms evaluated with it.", "scope": "A reinforcement-learning result is easier to interpret when the task, observations and reward have a consistent definition. The 2018 DeepMind Control Suite supplies a collection of continuous-control tasks, Python interfaces and baseline evaluations built on MuJoCo. It separates physical domains from task objectives and uses a common reward scale for most benchmark tasks. This makes it easier to compare learning curves and inspect failure cases. The contribution here is the original benchmark resource and interface design, not the invention of MuJoCo or the algorithms evaluated with it. Most default observations are state-derived features, with pixel observations available separately. Fixed 1,000-step returns are evaluation proxies, not proofs about infinite-horizon performance. Baselines use different numbers of seeds and parallel actors; their settings are starting points, not upper bounds for each algorithm. Stability checks are empirical. The later 2020 dm_control account adds software and tasks with a different byline; those additions are not silently included in this 2018 award.", "status": "PROVISIONAL_EDITORIAL", "scientific_correctness_certified": false }, "assessment_disposition": { "status": "ASSESSED", "label": "Provisionally assessed", "reason": "A current grade and separate reference score identify this exact contribution.", "target": { "record_id": "urn:uuid:e1b7d900-54eb-4b27-a863-fa1f093a6942", "version": 1, "content_hash": "sha256:496c87a73abad1835c64e604700fd1005ad5e0258c7da273f90f9d281de6d360" }, "edition": "ratings-e039106f30dec177" } }, "version": { "kind": "PUBLIC_RECORD_VIEW", "previous": null, "published_at": "2026-10-23T00:00:00Z", "publisher": { "principal_id": "urn:uuid:80116931-785b-40a2-a24b-9d59db0c7e33", "role": "WRS_EDITORIAL_PUBLICATION" }, "record": { "contributors": [ { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Yuval Tassa", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "DeepMind Control Suite", "url": "https://arxiv.org/abs/1801.00690v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Yotam Doron", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "DeepMind Control Suite", "url": "https://arxiv.org/abs/1801.00690v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Alistair Muldal", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "DeepMind Control Suite", "url": "https://arxiv.org/abs/1801.00690v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Tom Erez", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "DeepMind Control Suite", "url": "https://arxiv.org/abs/1801.00690v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Yazhe Li", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "DeepMind Control Suite", "url": "https://arxiv.org/abs/1801.00690v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Diego de Las Casas", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "DeepMind Control Suite", "url": "https://arxiv.org/abs/1801.00690v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "David Budden", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "DeepMind Control Suite", "url": "https://arxiv.org/abs/1801.00690v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Abbas Abdolmaleki", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "DeepMind Control Suite", "url": "https://arxiv.org/abs/1801.00690v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Josh Merel", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "DeepMind Control Suite", "url": "https://arxiv.org/abs/1801.00690v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Andrew Lefrancq", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "DeepMind Control Suite", "url": "https://arxiv.org/abs/1801.00690v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Timothy Lillicrap", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "DeepMind Control Suite", "url": "https://arxiv.org/abs/1801.00690v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Martin Riedmiller", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "DeepMind Control Suite", "url": "https://arxiv.org/abs/1801.00690v1" } } ], "evidence": [ { "description": "Bibliographic/stated-result source; not a WRS proof verification", "id": "source-0", "kind": "source_reference", "source": { "citation": "DeepMind Control Suite", "url": "https://arxiv.org/abs/1801.00690v1" } } ], "origin": "HUMAN_LEGACY", "record_profile": "wrs-local-research-notes-0.1", "record_type": "research_note", "rrp_version": "0.1", "scope": { "description": "Most default observations are state-derived features, with pixel observations available separately. Fixed 1,000-step returns are evaluation proxies, not proofs about infinite-horizon performance. Baselines use different numbers of seeds and parallel actors; their settings are starting points, not upper bounds for each algorithm. Stability checks are empirical. The later 2020 dm_control account adds software and tasks with a different byline; those additions are not silently included in this 2018 award.", "domain": "Continuous-control benchmark infrastructure" }, "sources": [ { "citation": "DeepMind Control Suite", "url": "https://arxiv.org/abs/1801.00690v1" } ], "statement": "# DeepMind Control Suite: comparable continuous-control tasks\n\nA reinforcement-learning result is easier to interpret when the task, observations and reward have a consistent definition. The 2018 DeepMind Control Suite supplies a collection of continuous-control tasks, Python interfaces and baseline evaluations built on MuJoCo. It separates physical domains from task objectives and uses a common reward scale for most benchmark tasks. This makes it easier to compare learning curves and inspect failure cases. The contribution here is the original benchmark resource and interface design, not the invention of MuJoCo or the algorithms evaluated with it.\n\n## Scientific scope\n\nMost default observations are state-derived features, with pixel observations available separately. Fixed 1,000-step returns are evaluation proxies, not proofs about infinite-horizon performance. Baselines use different numbers of seeds and parallel actors; their settings are starting points, not upper bounds for each algorithm. Stability checks are empirical. The later 2020 dm_control account adds software and tasks with a different byline; those additions are not silently included in this 2018 award.\n\n## Reading status\n\nThis is a source-linked WRS catalogue note. A full explanatory guide and independent scientific reconstruction have not been completed for this entry. Read the identified original work for its complete narrative and arguments.\n\n## People and systems\n\nYuval Tassa; Yotam Doron; Alistair Muldal; Tom Erez; Yazhe Li; Diego de Las Casas; David Budden; Abbas Abdolmaleki; Josh Merel; Andrew Lefrancq; Timothy Lillicrap; Martin Riedmiller\n\nSource-listed human researchers and engineers for the exact method or resource. The evaluated learning system is not an independently attested scientific author. Earlier methods, funding and data-collection roles remain separate. Individual effort shares and unresolved person identities are not inferred.\n\n\n\n## Original sources\n\n- [DeepMind Control Suite](https://arxiv.org/abs/1801.00690v1)\n", "summary": "A reinforcement-learning result is easier to interpret when the task, observations and reward have a consistent definition. The 2018 DeepMind Control Suite supplies a collection of continuous-control tasks, Python interfaces and baseline evaluations built on MuJoCo. It separates physical domains from task objectives and uses a common reward scale for most benchmark tasks. This makes it easier to compare learning curves and inspect failure cases. The contribution here is the original benchmark resource and interface design, not the invention of MuJoCo or the algorithms evaluated with it.", "title": "DeepMind Control Suite: comparable continuous-control tasks" }, "record_id": "urn:uuid:e1b7d900-54eb-4b27-a863-fa1f093a6942", "version": 1, "source_ref": { "record_id": "urn:uuid:e1b7d900-54eb-4b27-a863-fa1f093a6942", "version": 1, "content_hash": "sha256:496c87a73abad1835c64e604700fd1005ad5e0258c7da273f90f9d281de6d360" }, "export_view": "wrs-public-editorial-view-1", "public_view_hash": "sha256:a275534ab33fce7d076a04a6313c0eb7c690b78e5fdc9e62ac0a865427265f33" }, "target": { "record_id": "urn:uuid:e1b7d900-54eb-4b27-a863-fa1f093a6942", "version": 1, "content_hash": "sha256:496c87a73abad1835c64e604700fd1005ad5e0258c7da273f90f9d281de6d360" }, "published_grade": { "profile": "wrs-published-grade-1", "edition": "2026-10-28.1", "id": "tassa-doron-muldal-et-al-dmcontrol-task-software-2018", "entry_id": "tassa-doron-muldal-et-al-dmcontrol-task-software-2018", "revision": 1, "title": "DeepMind Control Suite: comparable continuous-control tasks", "context": "Continuous-control benchmark infrastructure", "track": "Reusable scientific method, experimental result or historical framework", "band": "A", "modifier": "+", "status": "PROVISIONAL_EDITORIAL", "assessed_on": "2026-10-23", "evidence_cutoff": "2026-10-23", "achievement_state": "SUPPORTED_SCOPED_RESULT", "what_changed": "A reinforcement-learning result is easier to interpret when the task, observations and reward have a consistent definition. The 2018 DeepMind Control Suite supplies a collection of continuous-control tasks, Python interfaces and baseline evaluations built on MuJoCo. It separates physical domains from task objectives and uses a common reward scale for most benchmark tasks. This makes it easier to compare learning curves and inspect failure cases. The contribution here is the original benchmark resource and interface design, not the invention of MuJoCo or the algorithms evaluated with it.", "scope": "A reinforcement-learning result is easier to interpret when the task, observations and reward have a consistent definition. The 2018 DeepMind Control Suite supplies a collection of continuous-control tasks, Python interfaces and baseline evaluations built on MuJoCo. It separates physical domains from task objectives and uses a common reward scale for most benchmark tasks. This makes it easier to compare learning curves and inspect failure cases. The contribution here is the original benchmark resource and interface design, not the invention of MuJoCo or the algorithms evaluated with it. Most default observations are state-derived features, with pixel observations available separately. Fixed 1,000-step returns are evaluation proxies, not proofs about infinite-horizon performance. Baselines use different numbers of seeds and parallel actors; their settings are starting points, not upper bounds for each algorithm. Stability checks are empirical. The later 2020 dm_control account adds software and tasks with a different byline; those additions are not silently included in this 2018 award.", "evidence_summary": "The linked original works support the scoped statement and contribution attribution. WRS has not independently replayed experiments or reconstructed a full mathematical proof. Most default observations are state-derived features, with pixel observations available separately. Fixed 1,000-step returns are evaluation proxies, not proofs about infinite-horizon performance. Baselines use different numbers of seeds and parallel actors; their settings are starting points, not upper bounds for each algorithm. Stability checks are empirical. The later 2020 dm_control account adds software and tasks with a different byline; those additions are not silently included in this 2018 award.", "references": [ { "title": "DeepMind Control Suite", "url": "https://arxiv.org/abs/1801.00690v1", "locator": "Original scoped contribution, statement/method and evidence limits; source-specific attribution retained." } ], "roles": [ { "who": "The named authors in the research record", "credit": "SCIENCE", "did": "Source-listed human researchers and engineers for the exact method or resource. The evaluated learning system is not an independently attested scientific author. Earlier methods, funding and data-collection roles remain separate. Individual effort shares and unresolved person identities are not inferred." } ], "target": { "record_id": "urn:uuid:e1b7d900-54eb-4b27-a863-fa1f093a6942", "version": 1, "content_hash": "sha256:496c87a73abad1835c64e604700fd1005ad5e0258c7da273f90f9d281de6d360" }, "scientific_correctness_certified": false, "independent_expert_calibration": false, "method": "PROPRIETARY", "points": "160000" }, "canonical_agent_submission": false, "scientific_verification_awarded": false, "reader_guide": null, "reading_status": "SOURCE_LINKED_NOTE" }