Investigating Compounding Prediction Errors in Learned Dynamics Models
Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 reserved pith:BHNLPCNArecord.jsonopen to challenge →
read the original abstract
Accurately predicting the consequences of agents' actions is a key prerequisite for planning in robotic control. Model-based reinforcement learning (MBRL) is one paradigm which relies on the iterative learning and prediction of state-action transitions to solve a task. Deep MBRL has become a popular candidate, using a neural network to learn a dynamics model that predicts with each pass from high-dimensional states to actions. These "one-step" predictions are known to become inaccurate over longer horizons of composed prediction - called the compounding error problem. Given the prevalence of the compounding error problem in MBRL and related fields of data-driven control, we set out to understand the properties of and conditions causing these long-horizon errors. In this paper, we explore the effects of subcomponents of a control problem on long term prediction error: including choosing a system, collecting data, and training a model. These detailed quantitative studies on simulated and real-world data show that the underlying dynamics of a system are the strongest factor determining the shape and magnitude of prediction error. Given a clearer understanding of compounding prediction error, researchers can implement new types of models beyond "one-step" that are more useful for control.
This paper has not been read by Pith yet.
Forward citations
Cited by 6 Pith papers
-
Long-Horizon Model-Based Offline Reinforcement Learning Without Explicit Conservatism
NEUBAY uses Bayesian posteriors over world models with long-horizon planning to match or exceed conservative offline RL methods without explicit conservatism.
-
Physics-Aware Sparse Learning and Selective Online Adaptation for Euler-Lagrange Robot Dynamics
A physics-aware residual framework decomposes Euler-Lagrange model mismatch into constrained inertia/Coriolis corrections plus a sparse history-dependent latent disturbance model adapted via Bayesian regression, impro...
-
A Definition and Roadmap for World Models
A perspective article defining world models as finite-resource compression of physical state transitions and outlining a roadmap toward physical AGI via unified representations and interactive simulators.
-
Scaling World-Model Reinforcement Learning Through Diffusion Policy Optimization
MBDPO reformulates policy optimization as a diffusion process over searched trajectories in latent world models to reduce misalignment between search and value learning.
-
Learning Cross-Coupled and Regime Dependent Dynamics for Aerial Manipulation
A nonlinear latent encoder plus linear latent decoder learns cross-coupled, history-dependent, and nonstationary residual dynamics for aerial manipulators, supporting online Bayesian adaptation and improved real-time ...
-
Intention-Conditioned Flow Occupancy Models
InFOM applies flow matching to model intention-conditioned occupancy measures for RL pre-training, reporting 1.8x median return gains and 36% higher success rates on benchmarks.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.