Causal world models under physical parameter shifts

Key points

• World models predict how embodied systems evolve under actions, but their accuracy can degrade when physical conditions move outside the training distribution.1
• We study two components of causal modeling: which inputs a predictor may use, and how its mechanism computes their effects.1
• In a shared 100-step forecast, the learned-torque model is competitive with the strongest recurrent robotic world model (RWM) configuration (median joint R2 of 0.755 - 0.768 versus 0.752), whereas supplying out-of-range masses sharply degrades the MLP and recurrent predictors.1
• These results indicate that how a physical parameter enters the dynamics mechanism determines whether its out-of-range value helps or harms prediction.1

Sources (1)

• [1]Causal world models under physical parameter shifts
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 21, 02:26 AM

  For robot link-mass extrapolation in CausalWorld, with disjoint training and test mass ranges, we develop finger-local world models in which link masses enter inertia and dynamics-bias terms through explicit affine coefficients, actions act through either a learned torque network or the known PD controller, and learned residuals capture contact-related effects.

  World models predict how embodied systems evolve under actions, but their accuracy can degrade when physical conditions move outside the training distribution.

Extractive summary: sentences quoted from the sources.