Mosim Humanoid Walk Random Ar1 Dt005: Humanoid Robot State Transitions

Humanoid Walk state transition data collected from the DMC (DeepMind Control Suite) environment using random AR(1)-style actions. The dataset contains trajectories of state-action-next_state tuples, with a state dimension of 55 and an action dimension of 21. It was authored by Weyl09 and last updated on 2026-05-22.

Use Cases

• Training reinforcement learning agents for bipedal locomotion based on state-action transitions.
• Benchmarking model-based control algorithms using the provided state and action dimensions.
• Analyzing the dynamics of humanoid walking based on the AR(1) action noise model mentioned in the description.

Strengths

• Dataset provides specific state and action dimensions (state_dim=55, action_dim=21).
• Data is structured as [s_t, a_t, s_(t+1)] transition rows, which is a standard format for RL.
• Trajectories are stored in a compressed .npz format suitable for efficient loading.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Freshness should be verified as the last update timestamp is from the future (2026-05-22).

Provenance

Source

huggingface

Collection Method

Collected from DMC (DeepMind Control Suite) with random AR(1)-style actions.

Freshness

Last updated 2026-05-22 05:01:23