Datasets:

KlaskLab
/

klasktron-il-benchmark

+{
+  "@context": {
+    "@language": "en",
+    "@vocab": "https://schema.org/",
+    "arrayShape": "cr:arrayShape",
+    "citeAs": "cr:citeAs",
+    "column": "cr:column",
+    "conformsTo": "dct:conformsTo",
+    "containedIn": "cr:containedIn",
+    "cr": "http://mlcommons.org/croissant/",
+    "data": {
+      "@id": "cr:data",
+      "@type": "@json"
+    },
+    "dataBiases": "cr:dataBiases",
+    "dataCollection": "cr:dataCollection",
+    "dataType": {
+      "@id": "cr:dataType",
+      "@type": "@vocab"
+    },
+    "dct": "http://purl.org/dc/terms/",
+    "extract": "cr:extract",
+    "field": "cr:field",
+    "fileProperty": "cr:fileProperty",
+    "fileObject": "cr:fileObject",
+    "fileSet": "cr:fileSet",
+    "format": "cr:format",
+    "includes": "cr:includes",
+    "isArray": "cr:isArray",
+    "isLiveDataset": "cr:isLiveDataset",
+    "jsonPath": "cr:jsonPath",
+    "key": "cr:key",
+    "md5": "cr:md5",
+    "parentField": "cr:parentField",
+    "path": "cr:path",
+    "personalSensitiveInformation": "cr:personalSensitiveInformation",
+    "recordSet": "cr:recordSet",
+    "references": "cr:references",
+    "regex": "cr:regex",
+    "repeated": "cr:repeated",
+    "replace": "cr:replace",
+    "sc": "https://schema.org/",
+    "separator": "cr:separator",
+    "source": "cr:source",
+    "subField": "cr:subField",
+    "transform": "cr:transform",
+    "rai": "http://mlcommons.org/croissant/RAI/",
+    "prov": "http://www.w3.org/ns/prov#"
+  },
+  "@type": "sc:Dataset",
+  "distribution": [
+    {
+      "@type": "cr:FileObject",
+      "@id": "repo",
+      "name": "repo",
+      "description": "The Hugging Face git repository.",
+      "contentUrl": "https://huggingface.co/datasets/KlaskLab/klasktron-il-benchmark/tree/refs%2Fconvert%2Fparquet",
+      "encodingFormat": "git+https",
+      "sha256": "https://github.com/mlcommons/croissant/issues/80"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "parquet-files-for-config-synthetic",
+      "containedIn": {
+        "@id": "repo"
+      },
+      "encodingFormat": "application/x-parquet",
+      "includes": "synthetic/*/*.parquet"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "parquet-files-for-config-human",
+      "containedIn": {
+        "@id": "repo"
+      },
+      "encodingFormat": "application/x-parquet",
+      "includes": "human/*/*.parquet"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "parquet-files-for-config-human_augmented",
+      "containedIn": {
+        "@id": "repo"
+      },
+      "encodingFormat": "application/x-parquet",
+      "includes": "human_augmented/*/*.parquet"
+    }
+  ],
+  "recordSet": [
+    {
+      "@type": "cr:RecordSet",
+      "dataType": "cr:Split",
+      "key": {
+        "@id": "synthetic_splits/split_name"
+      },
+      "@id": "synthetic_splits",
+      "name": "synthetic_splits",
+      "description": "Splits for the synthetic config.",
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "synthetic_splits/split_name",
+          "dataType": "sc:Text"
+        }
+      ],
+      "data": [
+        {
+          "synthetic_splits/split_name": "train"
+        }
+      ]
+    },
+    {
+      "@type": "cr:RecordSet",
+      "@id": "synthetic",
+      "description": "KlaskLab/klasktron-il-benchmark - 'synthetic' subset",
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "synthetic/split",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-synthetic"
+            },
+            "extract": {
+              "fileProperty": "fullpath"
+            },
+            "transform": {
+              "regex": "synthetic/(?:partial-)?(train)/.+parquet$"
+            }
+          },
+          "references": {
+            "field": {
+              "@id": "synthetic_splits/split_name"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "synthetic/tier",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-synthetic"
+            },
+            "extract": {
+              "column": "tier"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "synthetic/episode_id",
+          "dataType": "cr:Int32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-synthetic"
+            },
+            "extract": {
+              "column": "episode_id"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "synthetic/source_file",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-synthetic"
+            },
+            "extract": {
+              "column": "source_file"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "synthetic/episode_in_file",
+          "dataType": "cr:Int32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-synthetic"
+            },
+            "extract": {
+              "column": "episode_in_file"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "synthetic/step",
+          "dataType": "cr:Int32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-synthetic"
+            },
+            "extract": {
+              "column": "step"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "synthetic/obs",
+          "dataType": "cr:Float32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-synthetic"
+            },
+            "extract": {
+              "column": "obs"
+            }
+          },
+          "isArray": true,
+          "arrayShape": "12"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "synthetic/act",
+          "dataType": "cr:Float32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-synthetic"
+            },
+            "extract": {
+              "column": "act"
+            }
+          },
+          "isArray": true,
+          "arrayShape": "4"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "synthetic/next_obs",
+          "dataType": "cr:Float32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-synthetic"
+            },
+            "extract": {
+              "column": "next_obs"
+            }
+          },
+          "isArray": true,
+          "arrayShape": "12"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "synthetic/rew",
+          "dataType": "cr:Float32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-synthetic"
+            },
+            "extract": {
+              "column": "rew"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "synthetic/terminal",
+          "dataType": "sc:Boolean",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-synthetic"
+            },
+            "extract": {
+              "column": "terminal"
+            }
+          }
+        }
+      ]
+    },
+    {
+      "@type": "cr:RecordSet",
+      "dataType": "cr:Split",
+      "key": {
+        "@id": "human_splits/split_name"
+      },
+      "@id": "human_splits",
+      "name": "human_splits",
+      "description": "Splits for the human config.",
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "human_splits/split_name",
+          "dataType": "sc:Text"
+        }
+      ],
+      "data": [
+        {
+          "human_splits/split_name": "train"
+        }
+      ]
+    },
+    {
+      "@type": "cr:RecordSet",
+      "@id": "human",
+      "description": "KlaskLab/klasktron-il-benchmark - 'human' subset",
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "human/split",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-human"
+            },
+            "extract": {
+              "fileProperty": "fullpath"
+            },
+            "transform": {
+              "regex": "human/(?:partial-)?(train)/.+parquet$"
+            }
+          },
+          "references": {
+            "field": {
+              "@id": "human_splits/split_name"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "human/tier",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-human"
+            },
+            "extract": {
+              "column": "tier"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "human/episode_id",
+          "dataType": "cr:Int32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-human"
+            },
+            "extract": {
+              "column": "episode_id"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "human/source_file",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-human"
+            },
+            "extract": {
+              "column": "source_file"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "human/episode_in_file",
+          "dataType": "cr:Int32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-human"
+            },
+            "extract": {
+              "column": "episode_in_file"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "human/step",
+          "dataType": "cr:Int32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-human"
+            },
+            "extract": {
+              "column": "step"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "human/obs",
+          "dataType": "cr:Float32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-human"
+            },
+            "extract": {
+              "column": "obs"
+            }
+          },
+          "isArray": true,
+          "arrayShape": "12"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "human/act",
+          "dataType": "cr:Float32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-human"
+            },
+            "extract": {
+              "column": "act"
+            }
+          },
+          "isArray": true,
+          "arrayShape": "4"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "human/next_obs",
+          "dataType": "cr:Float32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-human"
+            },
+            "extract": {
+              "column": "next_obs"
+            }
+          },
+          "isArray": true,
+          "arrayShape": "12"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "human/rew",
+          "dataType": "cr:Float32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-human"
+            },
+            "extract": {
+              "column": "rew"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "human/terminal",
+          "dataType": "sc:Boolean",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-human"
+            },
+            "extract": {
+              "column": "terminal"
+            }
+          }
+        }
+      ]
+    },
+    {
+      "@type": "cr:RecordSet",
+      "dataType": "cr:Split",
+      "key": {
+        "@id": "human_augmented_splits/split_name"
+      },
+      "@id": "human_augmented_splits",
+      "name": "human_augmented_splits",
+      "description": "Splits for the human_augmented config.",
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "human_augmented_splits/split_name",
+          "dataType": "sc:Text"
+        }
+      ],
+      "data": [
+        {
+          "human_augmented_splits/split_name": "train"
+        }
+      ]
+    },
+    {
+      "@type": "cr:RecordSet",
+      "@id": "human_augmented",
+      "description": "KlaskLab/klasktron-il-benchmark - 'human_augmented' subset",
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "human_augmented/split",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-human_augmented"
+            },
+            "extract": {
+              "fileProperty": "fullpath"
+            },
+            "transform": {
+              "regex": "human_augmented/(?:partial-)?(train)/.+parquet$"
+            }
+          },
+          "references": {
+            "field": {
+              "@id": "human_augmented_splits/split_name"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "human_augmented/tier",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-human_augmented"
+            },
+            "extract": {
+              "column": "tier"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "human_augmented/episode_id",
+          "dataType": "cr:Int32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-human_augmented"
+            },
+            "extract": {
+              "column": "episode_id"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "human_augmented/source_file",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-human_augmented"
+            },
+            "extract": {
+              "column": "source_file"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "human_augmented/episode_in_file",
+          "dataType": "cr:Int32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-human_augmented"
+            },
+            "extract": {
+              "column": "episode_in_file"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "human_augmented/step",
+          "dataType": "cr:Int32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-human_augmented"
+            },
+            "extract": {
+              "column": "step"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "human_augmented/obs",
+          "dataType": "cr:Float32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-human_augmented"
+            },
+            "extract": {
+              "column": "obs"
+            }
+          },
+          "isArray": true,
+          "arrayShape": "12"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "human_augmented/act",
+          "dataType": "cr:Float32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-human_augmented"
+            },
+            "extract": {
+              "column": "act"
+            }
+          },
+          "isArray": true,
+          "arrayShape": "4"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "human_augmented/next_obs",
+          "dataType": "cr:Float32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-human_augmented"
+            },
+            "extract": {
+              "column": "next_obs"
+            }
+          },
+          "isArray": true,
+          "arrayShape": "12"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "human_augmented/rew",
+          "dataType": "cr:Float32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-human_augmented"
+            },
+            "extract": {
+              "column": "rew"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "human_augmented/terminal",
+          "dataType": "sc:Boolean",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-human_augmented"
+            },
+            "extract": {
+              "column": "terminal"
+            }
+          }
+        }
+      ]
+    }
+  ],
+  "conformsTo": "http://mlcommons.org/croissant/1.1",
+  "name": "klasktron-il-benchmark",
+  "description": "\n\t\n\t\t\n\t\tKlaskTron\n\t\n\nKlaskTron is a Klask imitation-learning dataset with three tiers: real human play, axis-mirror-augmented human data, and large-scale synthetic rollouts.\n\n\t\n\t\t\nTier\nEpisodes\nSteps\nSource\n\n\n\t\t\nhuman\n111\n80,591\nreconstructed from recorded human play\n\n\nhuman_augmented\n444\n322,364\naxis-mirror augmentation of the human tier (none, x, y, xy)\n\n\nsynthetic\n34,649\n5,000,000\ngenerated in simulation from an expert-like policy\n\n\n\t\n\nTotal: 35,204 episodes and 5,402,955 action steps.… See the full description on the dataset page: https://huggingface.co/datasets/KlaskLab/klasktron-il-benchmark.",
+  "alternateName": [
+    "KlaskLab/klasktron-il-benchmark",
+    "KlaskTron"
+  ],
+  "creator": {
+    "@type": "Person",
+    "name": "KlaskLab",
+    "url": "https://huggingface.co/KlaskLab"
+  },
+  "keywords": [
+    "robotics",
+    "reinforcement-learning",
+    "original",
+    "cc-by-4.0",
+    "1M - 10M",
+    "parquet",
+    "Tabular",
+    "Text",
+    "Time-series",
+    "Datasets",
+    "Dask",
+    "Polars",
+    "Croissant",
+    "🇺🇸 Region: US",
+    "imitation-learning",
+    "reinforcement-learning",
+    "robotics",
+    "control",
+    "simulation",
+    "klask",
+    "parquet",
+    "pickle",
+    "tabular",
+    "datasets",
+    "mlcroissant"
+  ],
+  "license": "https://choosealicense.com/licenses/cc-by-4.0/",
+  "url": "https://huggingface.co/datasets/KlaskLab/klasktron-il-benchmark",
+  "rai:dataLimitations": "The benchmark exposes a 12D state vector only; no image observations are included. Magnetic biscuit obstacles are omitted, isolating contact-rich and adversarial dynamics from combinatorial rule complexity. The BC scaling study uses a single run per dataset scale, so the non-monotonic trend should be treated as a diagnostic observation rather than a statistically confirmed scaling law. Sim-to-real transfer is validated on a single physical hardware instance only.",
+  "rai:dataBiases": "Tier 1 demonstrates recreational-to-intermediate skill level only and does not represent expert Klask play. Action labels in Tier 1 are reconstructed via finite differencing, making them nearly deterministically predictable from consecutive observations (data leakage risk for methods consuming velocity state components). The synthetic expert (Elo 1283, 61.8% win rate vs. 1200 baseline) exhibits systematic targeting errors of 0.5-1.5 cm during offensive maneuvers and does not represent ceiling performance. All data is collected from a single physical Klask board; cross-board hardware variability is not characterized.",
+  "rai:personalSensitiveInformation": "None. No personally identifying information is collected or released. Player identity is not recorded. The dataset contains only board state vectors (positions and velocities of pegs and ball) and action labels.",
+  "rai:dataUseCases": "The dataset is intended to measure imitation learning performance under contact-rich, adversarial dynamics — specifically covariate shift induced by an active opponent. Validated use cases: (1) behavioral cloning scaling studies (validated via BC trained on all four synthetic tiers, Section 5.2); (2) adversarial reward learning via GAIL (validated in Section 5.3); (3) demonstration-prior interaction with reward density (validated via 2x2 BC+RL ablation, Section 5.4). Not yet validated: offline RL on mixed human+synthetic corpora, visuomotor policy learning (no image observations included), and full-rules Klask including biscuit obstacles.",
+  "rai:dataSocialImpact": "Positive: provides the research community with an open, reproducible benchmark for adversarial imitation learning with a validated sim-to-real bridge, lowering the barrier to physical robot learning research. Negative: progress on adversarial IL enabled by this benchmark could in principle transfer to higher-stakes adversarial settings. However, the techniques studied (BC, GAIL, BC-bootstrapped PPO) are well-established and not differentially advanced by access to this benchmark. No fairness implications for specific communities are identified. The dataset contains no PII and was collected under informed consent. No access restrictions are imposed; the dataset is released under CC-BY-4.0.",
+  "rai:hasSyntheticData": true,
+  "prov:wasDerivedFrom": [
+    {
+      "@id": "https://huggingface.co/datasets/KlaskLab/klasktron-il-benchmark",
+      "prov:label": "klasktron-il-benchmark",
+      "sc:license": "CC-BY-4.0"
+    }
+  ],
+  "prov:wasGeneratedBy": [
+    {
+      "@type": "prov:Activity",
+      "prov:type": {
+        "@id": "https://www.wikidata.org/wiki/Q4929239"
+      },
+      "prov:label": "Human gameplay recording",
+      "sc:description": "Recorded during a single laboratory Klask tournament at 50 Hz via overhead camera with EKF state estimation. Actions reconstructed via finite differencing."
+    },
+    {
+      "@type": "prov:Activity",
+      "prov:type": {
+        "@id": "https://www.wikidata.org/wiki/Q5227332"
+      },
+      "prov:label": "Symmetry augmentation",
+      "sc:description": "ier 1 trajectories mirrored across x, y, and both axes to produce Tier 2 (4x augmentation)."
+    },
+    {
+      "@type": "prov:Activity",
+      "prov:type": {
+        "@id": "https://www.wikidata.org/wiki/Q4929239"
+      },
+      "prov:label": "Synthetic rollout generation",
+      "sc:description": "PPO expert policy rolled out for 5M timesteps in NVIDIA Isaac Lab via curriculum-based self-play."
+    }
+  ]
+}