Time Series Trajectories

Constructed in state-action-reward sequence form, supporting spatiotemporal understanding of models like Transformers.

Format: (s₀, a₀, [r₀]), (s₁, a₁, [r₁]), …, (s_T, a_T, [r_T])

State-Action Pair Collections

Basic data format for Behavior Cloning (BC) methods, containing expert demonstration state-action pairs (s, a).

Offline Reinforcement Learning Data