387 items · Page 1 of 9
Create a free account to keep your reading organized, join thoughtful discussions, and get more from every chapter.
Free to join · Takes less than a minute
Why
I created this space so readers can learn together, ask questions, and make sense of difficult ideas.
Model legged locomotion, whole-body control, and map-based navigation with hybrid contact dynamics, terrain uncertainty, and sim-to-real transfer.
Build a miniature planar pushing world model in Python: contact features, Ridge regressors, rollout error, random-shooting plans, and affordance maps.
Compare Atari, board games, and open worlds as learned game engines. Measure one-step prediction, recursive rollout drift, and model-predictive control.
VLA and world-action models fuse vision, language, and control, tokenize robot actions, and transfer across embodiments under closed-loop evaluation.
Examines NVIDIA's Cosmos world foundation models: video tokenization, physical AI data curation, conditional generation, guidance calibration, and simulation.
Explains how DINO-WM freezes a pretrained visual encoder, trains action-conditioned feature dynamics, and uses goal-image planning.
Explains how fixed datasets limit offline RL, why model exploitation can mislead planners, and how pessimism and policy constraints reduce unsupported choices.
Explains how TD-MPC pairs short-horizon latent planning with a learned terminal value, why decoder-free control-centric representations help, and its limits.
Explains how MuZero plans with learned latent dynamics and no reconstruction loss, plus a tutorial probing value equivalence and search targets.
Explains how Dreamer learns behaviors by latent imagination: RSSM dynamics, actor-critic training, and the V1, V2, and V3 design changes.
Compare World Models, SimPLe, and PlaNet: latent dynamics, video prediction, and CEM planning for model-based reinforcement learning.
Compare PILCO, PETS, and MBPO for decision-centric model-based reinforcement learning, covering deep ensembles, uncertainty, planning, and policy optimization.
Explains how options, symbolic STRIPS planning, language grounding, and multi-agent belief models structure long-horizon planning and where abstractions fail.
Explains how agents trade reward for information using dual control, information gain, curiosity, and safe exploration in world-model decision making.
Explains how actor-critic policies train inside learned world models, covering latent rollouts, lambda-returns, value expansion, and model exploitation.
Covers MCTS, UCT, PUCT, belief-space planning, and POMCP for decision-making under uncertainty with a hidden-state maze and tiger problem.
Explains how gradients flow through world-model rollouts, from adjoint backpropagation to action optimization, terminal values, and long-horizon stability.
Explains how sampling-based planning and model predictive control turn black-box world models into controllers, using random shooting, CEM.
Explains how world models are pretrained on diverse data, post-trained with preference objectives.
Explains how world models adapt over time through offline, online, and continual learning, covering replay, catastrophic forgetting, and safe drift guardrails.
Curation determines what a world model can learn. Examines deduplication, filtering, balancing, synthetic data, mixture weights, and contamination risks.
Explains how interaction data and passive observation shape world models, covering coverage, behavior policies, interventions, and data mixing strategies.
World models define and learn actions through inverse dynamics, controllability, and latent variables discovered in unlabeled video and aligned to controls.
Compare reconstruction, contrastive, masked, and multi-step objectives, including how each loss shapes rollout fidelity and control-relevant state.
Examines hierarchical, hybrid, and omnimodal world models, covering temporal abstraction, neural-symbolic hybrids, multimodal fusion.
Explains how JEPA predicts in latent space instead of pixels, why joint-embedding objectives collapse.
Explains how diffusion and flow world models represent multimodal futures via conditional denoising, trajectory diffusion, flow matching, guidance, and latency.
Explains how causal transformers predict world dynamics from tokenized observations and actions.
Compare attention-based transformers, structured state-space models, and hybrid architectures for world models.
Explains how recurrent state-space models combine deterministic memory with stochastic latents for filtering and imagination, and train them with the ELBO.
Explains how compositional world models reuse mechanisms, objects, and skills to generalize systematically to novel combinations.
Aleatoric and epistemic uncertainty shape world models, from multimodal transitions and ensembles to calibration and uncertainty propagation through rollouts.
Differentiable simulators, graph networks, neural ODEs, and operator learning add physical structure to world models for efficient data use and planning.
Covers structural causal models, the do-operator, counterfactuals, confounding, and invariance.
Embed physics into world models using Lagrangian and Hamiltonian architectures, contact and friction constraints, conservation laws.
Explains how coordinate frames, depth estimation, SLAM, occupancy grids, and neural fields give world models geometric state that stays stable under egomotion.
Explains how world models represent the agent itself: agent-environment boundaries, learned body schemas, capabilities, affordances.
Explains how object-centric and relational world models use slots, graphs, and message passing to predict object dynamics, track identity.
Explains how world models maintain state under partial observability using recurrent memory, gated LSTMs and GRUs, attention, episodic storage.
Explains what a world model's state must preserve and discard: predictive versus control sufficiency, bisimulation, minimal abstractions.
Explains how sensors shape world models: calibration, synchronization, noise, missing and delayed observations.
Explains how learned world models power planning, policy optimization, and value learning in RL, plus model bias, rollout error.
Covers control as inference, optimality variables, soft Bellman updates, variational free energy, active inference, preferences, ambiguity, and epistemic value.
Covers classical planning and optimal control: state-space search with A*, dynamic programming, LQR.
Covers system identification through excitation, regression, prediction-error and subspace methods, neural dynamics, rollout tests, and model validation.
Bayesian filters turn noisy observations into belief states. Covers Kalman, extended Kalman, unscented Kalman, particle, and learned update methods.
Explains how MDPs, POMDPs, policies, values, and belief states formalize sequential decisions when dynamics are uncertain and observations hide the true state.
Covers math behind world models: conditional distributions, Markov transitions, stability, attractors, and information theory for sufficient states.
Create a free account to keep your reading organized, join thoughtful discussions, and get more from every chapter.
Free to join · Takes less than a minute
Why
I created this space so readers can learn together, ask questions, and make sense of difficult ideas.