trajectories

Definition

Sequences of states over time. In reinforcement learning a trajectory is one rollout — the state, action, and reward tuples produced by running a policy to termination — and it is the unit of training data. In motion forecasting or autonomous driving it is the path an agent traces through space. The point of the word is that the whole sequence, not each timestep, is the object being modeled.

Explainers using this term

• Paper Walkthrough: Terminal-Universe — Turning Agent Logs Back Into Reusable Execution EnvironmentsTerminal-Universe: Turning Agent Trajectories into Scalable Terminal Environments
• Paper explained: DEFT-RLVR — show a driving VLM the future trajectory too early and it fabricates its reasoningDeferred Exposure of Future Trajectories for Verifiable Reasoning in Autonomous Driving VLMs