Skip to content
WORLD MODELS · TIMELINE

World Models: 36 Years from Internal Models to WAMs

1990–2026 · Five lineages · 33 milestones

Early explorationLatent-space eraVideo generation · Physical AI · VLAWAM unification1990200020102018202020222023202420252026Model-based RL · Latent DynamicsVideo-generation World ModelsJEPA Predictive EmbeddingsVLA · World Action ModelsEmbodied / Physical World ModelsWorld Models1990 · SchmidhuberPILCO2011 · Deisenroth & RasmussenWorld Models2018 · Ha & SchmidhuberPlaNet2018 · DeepMindMuZero2019 · DeepMindDreamer2019 · Hafner et al.DreamerV22020 · Hafner et al.TD-MPC2022 · Hansen et al.DreamerV32023 · Hafner et al.I-JEPA2023 · Meta AIGAIA-12023 · WayveUniSim2023 · GoogleSora2024 · OpenAIGenie2024 · DeepMindV-JEPA2024 · Meta AI1X World Model2024 · 1X TechnologiesGenie 22024 · DeepMindCosmos2025 · NVIDIAV-JEPA 22025 · Meta AIGenie 32025 · DeepMindRT-22023 · Google DeepMindOpenVLA2024 · Stanford et al.π02024 · Physical IntelligenceGR00T N12025 · NVIDIAHelix2025 · Figure AIWAMs Survey2026 · Fudan University · Shanghai Innovation Institute · NUSACT2023 · StanfordRDT / H-RDT2024 · Tsinghuaπ0.52025 · Physical IntelligenceOpenVLA-OFT2025 · Stanford · BerkeleyGR00T N1.72026 · NVIDIAXiaomi-Robotics-02026 · XiaomiGigaWorld-Policy-02026 · GigaWorld Team
The x-axis is non-linear (1990–2017 compressed). Hover a node for details.← swipe to explore the full timeline →
WAM PHYLOGENY · 2023–2026

Branching Evolution: From VLA to World Action Models

61 site-collected papers · four lineages · taxonomy per arXiv:2605.12090

Vertical axis is paper date (non-linear; the dense 2026 band is expanded). WAM unifies VLA's p(a|o,l) and world models' p(o′|o,a) into the joint distribution p(o′,a|o,l), splitting into cascaded (imagine, then act) and joint (one model generates both) families; the bottom band is the RoboDojo arena snapshot (Aug 2026, 34 models). Hover for details; click a node to open the paper.
PAPEROct 12, 2026

PACE: Temporal-Role Tactile Conditioning for Chunk-Based Insertion Policies

Contact-rich insertion requires a robot to combine visual alignment with tactile adjustment. Existing visuotactile chunk-based policies often collapse tactile feedback into one generation-time pathway, so persistent contact state, abrupt contact events, and the future contact consequences of an action are modeled implicitly in a single representation. PACE assigns one tactile stream three temporal roles. An exponential-smoothed contact state modulates action-chunk generation through layer-wise feature conditioning. A short-window tactile difference forms an event cue and drives a gated residual correction over only the first steps of the chunk. A training-only head predicts future tactile increments from decoder states as an auxiliary representation-learning objective. This preserves chunk-level action continuity while giving slowly varying contact phases and fast collision or jamming events separate pathways. Across five real insertion tasks, PACE reaches 90.7 percent average success, more than 44 percentage points above vision-only and single-path tactile-fusion baselines. It also averages 88.7 percent on six ManipulationNet geometry and clearance settings, including 0.1 mm clearance. Ablations show complementary contributions from Phase, Event, and Pred., while cross-signal and cross-architecture tests extend the principle to force/torque, joint current, and diffusion-policy backbones.

Zhiyuan Chen, Shixiong Xu, Shuai Chen

Tactile SensingRobot Manipulation
PAPEROct 11, 2026

P-BFM: Perceptive Behavioral Foundation Models for Humanoid Control via Unsupervised Reinforcement Learning

Behavioral foundation models (BFMs) provide humanoids with reusable repertoires of whole-body behaviors; in real-world humanoid control, the same behavioral intent may require fundamentally different motions under different terrain constraints. Covering such variation with terrain-matched demonstrations requires enumerating a combinatorial space of behaviors and terrain geometries. We introduce P-BFM, a perceptive behavioral foundation model that factorizes humanoid control into terrain-invariant intent and terrain-conditioned realization. A forward-backward representation maps motion references, target states, and reward functions into a shared intent space without terrain information, while a perceptive policy learns how to realize each intent under the surrounding terrain geometry. We further identify a complementary limitation of purely latent control: behavioral latents can preserve local motion patterns while losing precise global progress. We therefore introduce Explicit Root Instructions (ERI), which separates global transport (e.g., heading and planar velocity) from behavioral intent, while leaving local whole-body adaptation to the perceptive policy. This decomposition leads to a characteristic capability absent from conventional BFMs: one fixed intent induces different, terrain-appropriate behaviors, including changes in step height, stride, posture, foothold placement, and contacts, without terrain-specific demonstrations. Across challenging terrains, P-BFM improves motion fidelity, root-control precision, and perturbation recovery over strong baselines. A single frozen policy further supports motion tracking, goal reaching, reward inference, teleoperation, and terrain-aware recovery, and transfers zero-shot to a Unitree G1 using onboard depth sensing.

Xue Wang, Yitang Li, Yiming Zhang

HumanoidUnsupervised Reinforcement Learning

As an Amazon Associate, we earn from qualifying purchases.