{"id":"action-conditioned-world-model","name":"Predict transitions conditional on actions","shortName":"Predict transitions conditional on actions","role":"Change an action while holding state fixed and compare predicted transitions with a held-out environment.","description":"An action-conditioned world model predicts consequences from state and action inputs. A visually plausible continuation from observational video is not sufficient evidence that the model distinguishes interventions. Training requires action-linked transitions with a stated coordinate system and observation timing. One-step prediction error can be small while errors compound during repeated predicted steps, especially when a policy seeks states absent from training. V-JEPA 2-AC and Dreamer use different learned-state mechanisms; the worked scalar transition below isolates their shared action-dependence requirement without replicating either.","mechanism":["Retain state, executed action and next observation as one time-aligned transition.","Fit or inspect a predictor that receives the action explicitly.","Hold initial state fixed and compare alternative action predictions.","Evaluate one-step and multi-step errors separately on held-out trajectories and policy-induced states."],"learningObjective":"Change an action while holding state fixed and compare predicted transitions with a held-out environment.","engineeringExample":"In an authored environment x_next=x+2a, x=1 with action +0.5 yields 2 and action −0.5 yields 0. A model x_hat_next=x+1.8a predicts 1.9 and 0.1, giving absolute error 0.1 each. Repeating action +0.5 for five predicted steps from x=1 gives model state 5.5 versus true state 6: error grows to 0.5. This exact linear example demonstrates accumulation, not a universal growth law.","evidenceNotes":"Mode: worked example. Example gain is 2 m per normalized command per step; model gain is 1.8. No clipping, noise or hidden state. This is worked arithmetic, not a trained world-model experience. Numerical examples are authored calculations, except explicitly identified automated core fixtures. Primary sources support the scoped mechanism; no human specialist review or learner outcome is implied.","specs":[{"label":"Learning mode","value":"worked example","context":"The executable decoder has 423 parameters; worked extensions do not imply an additional simulator."},{"label":"Worked contract","value":"Change an action while holding state fixed and compare predicted transitions with a held-out environment.","context":"Example gain is 2 m per normalized command per step; model gain is 1.8. No clipping, noise or hidden state. This is worked arithmetic, not a trained world-model experience."}],"interfaces":[{"name":"Inputs and state","direction":"in","signal":"Retain state, executed action and next observation as one time-aligned transition."},{"name":"Result and check","direction":"out","signal":"Change an action while holding state fixed and compare predicted transitions with a held-out environment."}],"tradeoff":"Action data supports control-relevant prediction but is limited by collected interventions and embodiment. A policy can exploit errors beyond that support.","whyAI":"Distinguish predicting observations, choosing actions and controlling a changing environment. Change an action while holding state fixed and compare predicted transitions with a held-out environment.","misconception":"It is a mistake to conclude: The generated scene merely looks realistic.","check":{"question":"What distinguishes action-conditioned prediction from an action-free continuation?","options":["Changing only the action can change the predicted next state.","The generated scene merely looks realistic.","Every future state is already included in the input."],"answer":0,"explanation":"Changing only the action can change the predicted next state. Example gain is 2 m per normalized command per step; model gain is 1.8. No clipping, noise or hidden state. This is worked arithmetic, not a trained world-model experience."},"prerequisites":["video-latent-prediction","loss-optimizer","evaluation"],"related":["video-latent-prediction","loss-optimizer","evaluation","imagined-rollouts-policy-learning","vision-language-action-interface","diffusion-action-horizon"],"sources":[{"title":"V-JEPA 2","url":"https://arxiv.org/html/2506.09985v1"},{"title":"Mastering Diverse Domains through World Models","url":"https://arxiv.org/html/2301.04104v2"}],"level":"model","scale":"Mathematical, data and behavior space; logical dimensions are not physical distance","kind":"model","branch":"model-lifecycle-world-action","subbranch":"Predict transitions conditional on actions","workstream":"W11","visualFamily":"mathematical","labId":"library","labScope":"Model lifecycle studio: worked example. Example gain is 2 m per normalized command per step; model gain is 1.8. No clipping, noise or hidden state. This is worked arithmetic, not a trained world-model experience.","science":[{"title":"Predict transitions conditional on actions · worked relation","equation":"\\hat x_{t+1}=f_\\theta(\\hat x_t,a_t),\\qquad e_H=|\\hat x_{t+H}-x_{t+H}|","explanation":"In an authored environment x_next=x+2a, x=1 with action +0.5 yields 2 and action −0.5 yields 0. A model x_hat_next=x+1.8a predicts 1.9 and 0.1, giving absolute error 0.1 each. Repeating action +0.5 for five predicted steps from x=1 gives model state 5.5 versus true state 6: error grows to 0.5. This exact linear example demonstrates accumulation, not a universal growth law.","assumptions":"Example gain is 2 m per normalized command per step; model gain is 1.8. No clipping, noise or hidden state. This is worked arithmetic, not a trained world-model experience."}],"history":[]}