license: apache-2.0
task_categories:
- reinforcement-learning
- other
language:
- code
tags:
- minecraft
- expert-demonstrations
- skill-segmentation
- action-segmentation
- object-centric
pretty_name: Minecraft Skill Segmentation Dataset
size_categories:
- 1K<n<10K
Dataset Card for Minecraft Expert Skill Data
This dataset consists of expert demonstration trajectories from a Minecraft simulation environment. Each trajectory includes ground-truth skill segmentation annotations, enabling research into action segmentation, skill discovery, imitation learning, and reinforcement learning with temporally-structured data.
Dataset Details
Dataset Description
The Minecraft Skill Segmentation Dataset contains gameplay trajectories from an expert policy in the Minecraft environment. Each trajectory is labeled with ground-truth skill boundaries and skill identifiers, allowing users to train and evaluate models for temporal segmentation, behavior cloning, and skill-based representation learning.
- Curated by: [dami2106]
- License: Apache 2.0
- Language(s) (NLP): Not applicable (code/visual)
Dataset Sources
- Repository: https://github.com/dami2106/Minecraft-Skill-Data
Uses
Direct Use
This dataset is designed for use in:
- Training models to segment long-horizon behaviors into reusable skills.
- Evaluating action segmentation or hierarchical RL approaches.
- Studying object-centric or spatially grounded RL methods.
- Pretraining representations from visual expert data.
Out-of-Scope Use
- Language-based tasks (no natural language data is included).
- Real-world robotics (simulation-only data).
- Tasks requiring raw image pixels if they are not included in your setup.
Dataset Structure
Each data file includes:
- A sequence of states (e.g., pixel POV observations, PCA features).
- Skill labels marking where each skill begins and ends.
Example structure:
{
"pixel_obs": [...], // Raw visual observations (e.g., RGB frames)
"pca_features": [...], // Compressed feature vectors (e.g., from CNN or ResNet)
"groundTruth": [...], // Ground-truth skill segmentation labels
"mapping": { // Mapping metadata for skill ID -> groundTruth label
"0": "chop_tree",
"1": "craft_table",
"2": "mine_stone",
...
}
}
Dataset Creation
Curation Rationale
This dataset was created to support research in skill discovery and temporal abstraction in complex, open-ended environments like Minecraft. The environment supports high-level goals and diverse interactions, making it suitable for testing generalizable skills.
Source Data
Data Collection and Processing
- Expert trajectories were generated using a scripted or trained policy within the Minecraft simulation.
- Skill labels were added based on environment signals (e.g., changes to inventory, task completions, block state transitions) and verified using heuristics.
Who are the source data producers?
The data was generated programmatically in the Minecraft simulation environment by expert agents using scripted or learned behavior policies.
Annotations
Annotation process
Skill annotations were derived from internal game state events and heuristics related to player intent and task segmentation. Manual inspection was performed to ensure consistency across trajectories.
Who are the annotators?
Automated rule-based annotation systems with developer oversight during dataset development.
Bias, Risks, and Limitations
- The dataset is derived from simulation, so its findings may not generalize to real-world robotics or broader RL environments.
- Skill definitions depend on domain-specific heuristics, which may not reflect all valid strategies.
- Expert strategies may be biased toward specific pathways (e.g., speedrunning logic).
Recommendations
Researchers should evaluate the robustness of learned skills across diverse environments and initial conditions. Segmentations reflect task approximations and should be interpreted within the scope of the simulation constraints.